)]}'
{
  "log": [
    {
      "commit": "68b4581468e14804ef554a4154df31cfcc4e01fd",
      "tree": "5067d3c530811716e6983fef345d8b1eece6e188",
      "parents": [
        "74aec7cc2835d0928202b7ed60a7114ad4c09c4c"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Oct 19 16:30:46 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Oct 19 16:30:50 2026 -1000"
      },
      "message": "Refuse finbench inputs the service used to default or drop\n\nmodel, percentiles, prob_tolerance, check_positive_definite, asset_ids,\ncovariance_basis and scenarioAnalysis n_assets are now refused with a\nnamed error when wrong, instead of being replaced, truncated or dropped;\nuse_hash_lookup, which never changed anything, is refused. The MCP and\nservices.xml schemas now state the limits the code enforces.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "74aec7cc2835d0928202b7ed60a7114ad4c09c4c",
      "tree": "714fec8f9515d25a4731fd706477e827b079a999",
      "parents": [
        "72514ccd936cf03afae978dd8aa52b56b3000fae"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Oct 19 16:01:20 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Oct 19 16:01:20 2026 -1000"
      },
      "message": "Make camera sample stubs fail clearly and drop shell examples\n\nThe stubs returned success without doing anything, so a deployment that\nmissed one reported recordings and transfers that never happened; they now\nanswer \"not implemented\" naming the function, unless built with\n-DCAMERA_STUBS_SUCCEED. The commented examples ran commands through\nsystem(), open to shell injection, and disabled SSH host key checking.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "72514ccd936cf03afae978dd8aa52b56b3000fae",
      "tree": "ebf76d32a2e936b6e1a654b9781d3245c516881c",
      "parents": [
        "89baaa1b657e4f5e2444d0740fdc19911e436311"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Oct 16 13:54:31 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Oct 16 13:54:31 2026 -1000"
      },
      "message": "Write a non-finite upside_downside_ratio as JSON null\n\njson-c wrote the all-upside and no-movement cases as the bare tokens\nInfinity and NaN, which strict JSON parsers reject. The response now\ncarries null there; upside_potential and downside_risk beside it still\ntell the two cases apart.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "89baaa1b657e4f5e2444d0740fdc19911e436311",
      "tree": "5659f3a25fd304193e1205bbaca360c2d194356b",
      "parents": [
        "52680113be6b54d750d52c85e823c77f82a2ed09"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Oct 15 15:55:36 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Oct 15 15:55:36 2026 -1000"
      },
      "message": "Refuse a scenarioAnalysis asset without a positive position_size\n\nThe parser left an absent position_size at 0, so a request that\nlooked valid came back with every dollar figure zeroed and the\nupside/downside ratio NaN. It is now refused like a non-positive\ncurrent_price, and the MCP schema lists it as required.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "52680113be6b54d750d52c85e823c77f82a2ed09",
      "tree": "03c566137a1551019cf1cb795a99d1714656c207",
      "parents": [
        "55fa61653696f6457da8f273e14d26e84f5fa5e7"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Oct 15 15:35:03 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Oct 15 15:35:03 2026 -1000"
      },
      "message": "Route scenarioAnalysis in the finbench json_object dispatcher\n\nfinbench_dispatch_json_obj() had no branch for scenarioAnalysis, so on\nthe statically linked Android path every request to it was refused with\n\"Unknown action\" although services.xml and the metadata list it. It now\nroutes by name, and a request with no name and an \"assets\" array is\ninferred as scenarioAnalysis.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "55fa61653696f6457da8f273e14d26e84f5fa5e7",
      "tree": "5f4ca87d848b3ea39cbecaf10387c00d56a6ece2",
      "parents": [
        "7bff2e939deebd3c3c2f95a585f7267ae49ca678"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Oct 11 05:00:51 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Oct 11 05:00:51 2026 -1000"
      },
      "message": "Document registering mod_axis2 in a static httpd\u0027s modules.c\n\nA static httpd has no LoadModule, and configure does not know mod_axis2, so\naxis2_module must be added to modules.c by hand. Without it every /services\nrequest gets a 404 and never reaches Axis2/C.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "7bff2e939deebd3c3c2f95a585f7267ae49ca678",
      "tree": "7bb6791018f17aa062426bcb21242fc5bc291612",
      "parents": [
        "ce7c8f38176c2c9524ccbeb881d944f1940da35f"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Oct 11 04:45:59 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Oct 11 04:45:59 2026 -1000"
      },
      "message": "Document that JSON requests dispatch by operation name\n\nThe JSON path never sets doing_rest, so RESTLocation is not matched: a\ncatch-all is reached at /services/\u003csvc\u003e/\u003cop name\u003e, and the bare service URL\nanswers \"Operation Not Found\". The doc claimed RESTLocation drove dispatch.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "ce7c8f38176c2c9524ccbeb881d944f1940da35f",
      "tree": "419c490fd1fef7a6c895de67a2174ca2e61694e0",
      "parents": [
        "d95df9c7b873a3d1150bbef55dbabd5b99a1fd9f"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Oct 07 10:24:35 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Oct 07 10:24:35 2026 -1000"
      },
      "message": "Raise the documented OpenSSL minimum to 3.5.9\n\nOpenSSL 3.0 LTS reached end of life on 2026-09-07, so the 3.0.21 floor\nnamed a branch that no longer gets fixes. 3.5.9 is the current 3.5 LTS\nrelease and includes the September 2026 fixes. The CVE-check MIN map\nmoves with it.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "d95df9c7b873a3d1150bbef55dbabd5b99a1fd9f",
      "tree": "56189544fd3bb0b458e6aaa9430e2c833295356c",
      "parents": [
        "4a22da0eaef88281067ce349764f1d58492d6571"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Oct 07 10:16:32 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Oct 07 10:16:32 2026 -1000"
      },
      "message": "Raise the documented libxml2 minimum to 2.15.4\n\nCVE-2026-86140 (fixed in 2.15.4) failed the monthly CVE check. Its overflow\nsits in DTD validation, which Axis2/C never enables, but the floor moves\nanyway. Also records the September OpenSSL batch (3.0 unaffected, 3.5.9 is\nthe new preferred floor) and syncs the stale table in the root SECURITY.md.\n\nCo-Authored-By: Claude Opus 5.5 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "4a22da0eaef88281067ce349764f1d58492d6571",
      "tree": "e9b31b88eca40b6f20a6153991e5f812a9b10d50",
      "parents": [
        "1b942ad05fb0fe59c90cc24157817458c47c29ee"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:27:57 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:27:57 2026 -1000"
      },
      "message": "Record the fuzzing work deferred: AXIOM ownership, json-c coverage\n\nThe XML leaks left are AXIOM\u0027s ownership model, deferred because the HTTP/2\nJSON path refuses XML bodies and parses only local config with AXIOM. Also\nnoted: the json-c target links an uninstrumented library.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "1b942ad05fb0fe59c90cc24157817458c47c29ee",
      "tree": "20aa1806fc19dee8d34b38eb1f37bf68b95cbce5",
      "parents": [
        "43ab1482c2b691ce2121bf010bb7c169c27baa3a"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "message": "Run the fuzz targets in CI\n\nBuilds with fuzz/oss-fuzz/build.sh as OSS-Fuzz would and runs each target\nunder ASan: a minute per push or PR, ten in the weekly run. Leak detection is\noff for the XML parser only, which still has unowned namespaces.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "43ab1482c2b691ce2121bf010bb7c169c27baa3a",
      "tree": "b81d28ad1d8f272dedd8e1dec4e6645bed3df372",
      "parents": [
        "94c5602deb7a93d644609f2ed2a3b9c53551975c"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "message": "Fix two namespace leaks in the AXIOM stax builder\n\nA prefix declared twice on one element dropped the first namespace unreleased;\nthe first is now kept and the duplicate skipped. An undeclared element prefix\nleaked the prefix string on the error path. Both inputs join the seed corpus.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "94c5602deb7a93d644609f2ed2a3b9c53551975c",
      "tree": "cbfe305429cbf1261d85af26b567df4adee42cc1",
      "parents": [
        "6eac3be45a9f77a6a251f8c6b8657e6f12f5a39b"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "message": "Keep the element frame when Guththila rejects an end tag\n\nA mismatched end tag returned with its element frame already popped, so the\nframe and its tokens leaked. It is pushed back and freed with the stack. The\ninput that found it is added to the XML seed corpus.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "6eac3be45a9f77a6a251f8c6b8657e6f12f5a39b",
      "tree": "0e5422d1788bff3d9d6d39bb2c3e7860a56f9e1e",
      "parents": [
        "a510272179d686177a7a9438ca92b5e6802414b5"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 27 07:19:25 2026 -1000"
      },
      "message": "Leave the standalone server out of the OSS-Fuzz build\n\nWith --disable-shared the standalone HTTP server links several plugins that\neach export axis2_get_instance, so the link fails and, under bash -eu, ends\nthe build before any fuzz target exists. No target uses the server.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "a510272179d686177a7a9438ca92b5e6802414b5",
      "tree": "f9e3e14727f23ad5a704293187be8da7de286d60",
      "parents": [
        "a7f5e17883b5eebc7f2c7b1c1772e9255fa1d83e"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 18:09:32 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 18:09:32 2026 -1000"
      },
      "message": "Link composeCovariance from the JSON user guide\u0027s resources\n\nPoints at the README section and at the header doc comment, which holds\nthe operation\u0027s full contract, so readers can find the Σ \u003d D·R·D\nfront door for hypothetical regimes without searching the sample.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "a7f5e17883b5eebc7f2c7b1c1772e9255fa1d83e",
      "tree": "30e9ec49e59305d63d91944c15b39ce7035550cc",
      "parents": [
        "b1870473099f2a26bf5d67aec3d84b0e3c6d7eb2"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 14:09:18 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 14:09:18 2026 -1000"
      },
      "message": "Put a plain-language summary at the top of the multithreading note\n\nThreads shared naively would make the same request give different answers from run to run; the note now says so first, with why in plain terms, before the design that restores repeatability at the cost of changing today\u0027s numbers.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "b1870473099f2a26bf5d67aec3d84b0e3c6d7eb2",
      "tree": "b0d997a0897dedb778b82cfe1090855e8732d0a5",
      "parents": [
        "23e90d364c422d6a921d1091aaa00519396de8b3"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 14:00:59 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 14:00:59 2026 -1000"
      },
      "message": "Record a multithreaded Monte Carlo design for future consideration\n\nThe financial benchmark\u0027s Monte Carlo runs on one thread. The note gives\nmeasured single-thread numbers on an x86-64 VM and an Android phone, what\nthreads would gain on each, and a block-seeded design whose results do not\ndepend on the thread count, kept beside the sequential engine as reference.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "23e90d364c422d6a921d1091aaa00519396de8b3",
      "tree": "ac0d5c702d3611aaa5566b7059ee212b5c894093",
      "parents": [
        "3294fe23843d69785157ac2494a41b98c7428513"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 13:11:23 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 13:11:23 2026 -1000"
      },
      "message": "Find or create a service group context under one lock\n\naxis2_conf_ctx_fill_ctxs looked the context up and registered a new one\nunder two separate holds of the mutex, so two first requests for a group\ncould both create one and the loser was overwritten and never freed. The\nlook-again, create and register now happen under a single hold.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "3294fe23843d69785157ac2494a41b98c7428513",
      "tree": "14955df0c052446839df4f3717b85bd7034342d4",
      "parents": [
        "6fcae89de1f57578cda1068256f87e9cba29428d"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 12:59:08 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 12:59:08 2026 -1000"
      },
      "message": "Describe the fixed teardown crash and the poisoning technique\n\nThe ASan and gdb guide said a child teardown crash was tracked separately.\nIt is fixed in 6fcae89de; the guide now gives its symptom and how the\nper-request env was caught: poison the request\u0027s env instead of destroying\nits pool, since ASan cannot see into mmap-backed APR pools.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "6fcae89de1f57578cda1068256f87e9cba29428d",
      "tree": "7c02b1f2eee67752213e41964f3415b7046803c1",
      "parents": [
        "8f38f77b2e7ffa1184a6bbcc7353118461394c7a"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 12:58:54 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 12:58:54 2026 -1000"
      },
      "message": "Build registered contexts with the conf_ctx\u0027s own env\n\naxis2_conf_ctx_fill_ctxs created the service group context it registers\nwith the env of whichever request needed it first. Under mod_axis2 that env\nlives in the request\u0027s pool, and the context\u0027s hash tables keep it, so\nevery Apache child crashed in axutil_hash_free as it exited. The conf_ctx\nnow keeps its creating env and builds registered contexts with it.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "8f38f77b2e7ffa1184a6bbcc7353118461394c7a",
      "tree": "8b75efc73d7f8b240c4de47cccc449e97a82c4fd",
      "parents": [
        "f683b78e5a6b63db555cb040aa3b41874125bc0d"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 12:31:36 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 26 12:31:36 2026 -1000"
      },
      "message": "Document ASan and gdb with the JSON samples under httpd\n\nA private ASan build, a separate single-process httpd with libasan\npreloaded, and gdb for what ASan cannot show: proving LeakSanitizer ran,\nlive heap bytes for growth that is not a leak, catching a signal-less crash.\nIncludes the LoadFile the JSON receiver\u0027s dlsym(RTLD_DEFAULT) lookup needs.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "f683b78e5a6b63db555cb040aa3b41874125bc0d",
      "tree": "f4e1b14d58e69eed5efd31938f5e78488be6b144",
      "parents": [
        "f7e6b224df09f424cc31b9cc6dd4b848b72ef41b"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 05:22:14 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 05:22:14 2026 -1000"
      },
      "message": "Start the h2 client\u0027s connect deadline after DNS resolution\n\nThe deadline was taken before getaddrinfo, so a slow lookup used up\nconnect_timeout_ms and failed the connect at once, contrary to the\nheader. It is now set once the host has resolved.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "f7e6b224df09f424cc31b9cc6dd4b848b72ef41b",
      "tree": "b923238bf4b9a48742103ff5d06ef04ae9e4ab1e",
      "parents": [
        "b6993d43ffe9e3f04a1978184ef6aa6d7a7a5c8c"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 03:49:57 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 03:49:57 2026 -1000"
      },
      "message": "State that the h2 client\u0027s connect timeout starts after DNS resolution\n\ngetaddrinfo blocks under the system resolver\u0027s own timeout. The header,\ndocs and call site now say so, and name an IP literal host with\nverify_name as the way to a hard connect bound.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "b6993d43ffe9e3f04a1978184ef6aa6d7a7a5c8c",
      "tree": "6c23a53b4acc76776227a107d7cc9008a8389399",
      "parents": [
        "f92cc730034b0ec223372138c6b61e7927f61764"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 03:47:53 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 03:47:53 2026 -1000"
      },
      "message": "Move request_timeout_ms to the end of the h2 client options\n\nNew options are appended so existing fields keep their offsets. The\nheader now states that rule and that it does not remove the need to\nrebuild against the header the program links with.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "f92cc730034b0ec223372138c6b61e7927f61764",
      "tree": "f69185c73550d231d39bab6a8870afce2242558d",
      "parents": [
        "a0c52b3b0c5f9995cdc4fd8d1e42a8d6fb7053a8"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 03:27:04 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 25 03:27:04 2026 -1000"
      },
      "message": "Correct the h2 client comment on a server that stops reading\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "a0c52b3b0c5f9995cdc4fd8d1e42a8d6fb7053a8",
      "tree": "4a56e5dca4691422d815f54e48ace827d29b9606",
      "parents": [
        "ca03d215d96ab94f73bf2d6f1d0a27ad6040d339"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 15:38:28 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 15:38:28 2026 -1000"
      },
      "message": "Note that a long-running h2 client call needs both timeouts raised\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "ca03d215d96ab94f73bf2d6f1d0a27ad6040d339",
      "tree": "36561182b5f5278aa2f6b548cfc6fd74f16f8bd5",
      "parents": [
        "46d843d1d9b159497a9cfcfaa9a0a798953e8e66"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 15:34:21 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 15:34:21 2026 -1000"
      },
      "message": "Bound a whole h2 client post with request_timeout_ms\n\nio_timeout_ms limits each wait, so a server trickling bytes could hold a\npost indefinitely. The new option (default 120 s) caps every wait at the\npost\u0027s overall deadline. Connecting stays under connect_timeout_ms.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "46d843d1d9b159497a9cfcfaa9a0a798953e8e66",
      "tree": "3e66c3403b70b24efd7f91144b3f8fd2b5d4bfd3",
      "parents": [
        "cac64816b243a2829874c9c1f69db8deb27c850e"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 14:12:28 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 14:12:28 2026 -1000"
      },
      "message": "Keep a complete h2 response when the server closes right after it\n\nA close_notify read in the same batch as the end of the response, or a\nfailed send of the trailing ACKs, discarded a response that had fully\narrived. Both now return it and drop the connection. content-length is\nformatted as unsigned long long, which holds any size_t.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "cac64816b243a2829874c9c1f69db8deb27c850e",
      "tree": "eb6dc8ac269141dd82f0ef1e09e35b01cf60310a",
      "parents": [
        "95fff174abb608ad0d0198c1affedd7106c4a4ef"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 13:22:28 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 24 13:22:28 2026 -1000"
      },
      "message": "Bound the h2 client\u0027s handshake writes by connect_timeout_ms\n\nA handshake write that blocked waited io_timeout_ms, not the remaining\nconnect deadline. Also: monotonic time is 64-bit (a 32-bit long of ms\nwraps after 24.8 days), response-buffer doubling cannot overflow size_t,\nand the file gains comments on how bytes flow through TLS and nghttp2.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "95fff174abb608ad0d0198c1affedd7106c4a4ef",
      "tree": "d783f2478bdf7e15729b13704ac0d4aad13bd1a1",
      "parents": [
        "b98ad331618a24159547b1a5a9265f1f3c3bb493"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 16:30:40 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 16:30:40 2026 -1000"
      },
      "message": "Reopen an HTTP/2 connection the server closed while it was idle\n\nhttpd closes idle HTTP/2 connections after its keep-alive timeout, and the\nclient found out only by failing the next post. Before sending on a reused\nconnection it now checks, without waiting, for a GOAWAY or FIN and opens\na new connection; nothing has been sent, so this is not a retry.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "b98ad331618a24159547b1a5a9265f1f3c3bb493",
      "tree": "4de8af4fd15ddb2a2841dde7cf3d712536fb7935",
      "parents": [
        "cfb435fa54dd1ad767644e6568c075833130c707"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 15:54:07 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 15:54:07 2026 -1000"
      },
      "message": "Add an HTTP/2 JSON client, libaxis2_h2_client\n\nAxis2/C could serve JSON over HTTP/2 but not call it. The new client POSTs\nJSON to a service operation over nghttp2 and OpenSSL, requires ALPN h2,\nalways verifies the server name, and supports mTLS. It needs only axutil,\nnghttp2 and OpenSSL, so embedded and Android callers link just that.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "cfb435fa54dd1ad767644e6568c075833130c707",
      "tree": "02280c87ec8ec6f4b0cab503bbdc9e0b5d629819",
      "parents": [
        "373f5cea900dc78cd8dbaf6a9a09b480b3a317b4"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 14:20:37 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 14:20:37 2026 -1000"
      },
      "message": "Read n_periods_per_year the same strict way in every operation\n\nA string, zero, a negative or a non-number is now refused instead of\nfalling back to the default, and portfolioVariance and monteCarlo also\nrefuse a fraction rather than truncate it. covarianceFromReturns still\ntakes a fractional factor, since it scales by it as a double.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "373f5cea900dc78cd8dbaf6a9a09b480b3a317b4",
      "tree": "5f8a9d4a88fcc8032a6a674297c0bdd537474777",
      "parents": [
        "1393ea33c74a9f6cc7ee1c338df4ddb47d2c36b8"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 13:56:35 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 13:56:35 2026 -1000"
      },
      "message": "Refuse an annualized matrix labelled with n_periods_per_year \u003e 1\n\nA covarianceFromReturns response passed back to portfolioVariance carries\nthe factor it already applied as n_periods_per_year beside\ncovariance_basis \"annualized\", which annualized the matrix twice.\nThat contradiction is now refused instead of guessed.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "1393ea33c74a9f6cc7ee1c338df4ddb47d2c36b8",
      "tree": "e29debb33ab081cf5fc5926eadbd42c06aae365a",
      "parents": [
        "410f5251680f1bcf29ae87a833821ea40b2f3e34"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 13:26:53 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 22 13:26:53 2026 -1000"
      },
      "message": "Check the diagonal for finiteness in portfolioVariance\n\nThe symmetry loop skipped i \u003d\u003d j, so a NaN or -Inf on the diagonal\nbypassed the PSD guard and returned SUCCESS. The loop now includes the\ndiagonal, and a non-finite w\u0027Sigma*w (NaN weight, overflow) is refused.\n\nCo-Authored-By: Claude Opus 5.5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "410f5251680f1bcf29ae87a833821ea40b2f3e34",
      "tree": "e87233f934ff4dd404ff8ca02c74fbabcb3bee49",
      "parents": [
        "22274a4218c16eb38730829a7397c752f7deee96"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 20:28:48 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 20:29:07 2026 -1000"
      },
      "message": "Add the financial-benchmark fuzzer to the OSS-Fuzz build with seeds\n\nfuzz_finbench existed but was never built by fuzz/oss-fuzz/build.sh, so\nthe service\u0027s request-to-response path ran only when someone fuzzed it\nby hand. It is now a target, compiled with the service source and linked\nlike the JSON reader fuzzer, with a 19-file seed corpus covering a valid\nand a deliberately invalid request per operation and a key dictionary.\nLocal libFuzzer + ASan run: 615K inputs in 90 s, 729 edges, no crashes.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "22274a4218c16eb38730829a7397c752f7deee96",
      "tree": "3239aedc0c38e8e1497ba285f932a08186aa58b6",
      "parents": [
        "d5864fd66fcd42884ae85a56a9f4eed7ddead473"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 20:15:42 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 20:15:42 2026 -1000"
      },
      "message": "Refuse asymmetric and non-PSD matrices in portfolioVariance\n\nSymmetry is now checked with the same tolerance monteCarlo and\ncomposeCovariance use, and a w\u0027Sigma*w that comes out negative beyond\nrounding scale is refused instead of being clamped to a zero volatility.\nThe in-function fallback for n_periods_per_year is 1, matching the parser\ndefault set earlier, so a zeroed struct never reverts to 252.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "d5864fd66fcd42884ae85a56a9f4eed7ddead473",
      "tree": "d1220ca5142cef570217a838c4078d5d068c7a38",
      "parents": [
        "eca314fb69063b0f45980598a572b347d84cbeca"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 20:02:41 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 20:02:41 2026 -1000"
      },
      "message": "Default portfolioVariance to an annualized matrix and echo the basis\n\nn_periods_per_year now defaults to 1, the basis composeCovariance,\ncovarianceFromReturns and every documented example produce; a caller with\na per-period matrix passes 252, 52 or 12. The response echoes\nn_periods_per_year and covariance_basis, and both producers stamp\ncovariance_basis \"annualized\". With the old 252 default an annualized\nmatrix reported annualized_volatility sqrt(252) too large, undetectably.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "eca314fb69063b0f45980598a572b347d84cbeca",
      "tree": "fdc682ee8f130f3ebb82e3a858d2304c0222d363",
      "parents": [
        "0446624d3e5577a986852278da1cf314d1732f4c"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 12:30:20 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 12:30:20 2026 -1000"
      },
      "message": "Let an embedding application add MCP tools to the sample transport\n\ntools/list and tools/call now consult finbench_mcp_extra_tools() and\nfinbench_mcp_extra_dispatch(), both weak and empty by default, so a plain\nbuild is unchanged. Built-ins are tried first, so an extra tool cannot\nshadow an operation. The sample has no data of its own; an application\nthat has some can keep the file reading and its path validation on its own\nside and still reach these operations through one transport.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "0446624d3e5577a986852278da1cf314d1732f4c",
      "tree": "2b62d00ca5d51ad8a520cf6579c7c63cfe2c56ff",
      "parents": [
        "5e4ac649c7a5d821a6f6d0b24d39f974e57c49fd"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 12:26:26 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 12:26:26 2026 -1000"
      },
      "message": "Add covarianceFromReturns to the financial benchmark sample\n\nSample covariance of a return matrix, Bessel-corrected and annualised by\nn_periods_per_year, producing the flat row-major matrix portfolioVariance\nand the correlated monteCarlo already take. Missing observations are\ncomplete-case and observations_per_pair reports what each pair would have\nhad, so thin data is visible rather than silently averaged over. Capped at\n50 assets and 5000 observations because the request itself is O(n*m).\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "5e4ac649c7a5d821a6f6d0b24d39f974e57c49fd",
      "tree": "c2944841a9d694d19c38d98fe80bd9f96290e568",
      "parents": [
        "09fa2a1e765896b399146745f6a9121f39c13ec4"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:52:43 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:52:43 2026 -1000"
      },
      "message": "Fail closed and escape correctly in the operation trust guard\n\nOnly an explicit RESTLocation of \"/\" is a catch-all now; an operation with\nno location parameter was resolved by its own name in the URL, so the body\nmay not override it. The refusal is built with json-c instead of a\nhand-formatted string, so a quote in the body cannot break or extend the\nresponse and nothing is truncated, and both names are sanitized before\nthey reach the log.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "09fa2a1e765896b399146745f6a9121f39c13ec4",
      "tree": "f718b2b36d0a1975e3b33e6e511b66f62fa26345",
      "parents": [
        "448d607f37a233a1fed0367dc7700f8ca4c13039"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:30:34 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:30:34 2026 -1000"
      },
      "message": "Let the Android cross-build finish: skip host-only codegen and .so validation\n\nConfigured for Android, the tree built every library and then failed\ntwice at the end: tools/codegen/native links libxml2, which the Android\nprefix does not carry, and the mod_axis2 install hook validates a shared\nmodule that a static-only build never produces. Both are now skipped\nunder the BUILD_FOR_ANDROID conditional, so build_android.sh followed by\nmake install exits 0. Documented in HTTP2_ANDROID.md.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "448d607f37a233a1fed0367dc7700f8ca4c13039",
      "tree": "a49cd1e829358a74b144c02c2290290dd7349a90",
      "parents": [
        "8a78396eb4568b26e10319589aa84d18da05108f"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:26:55 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:26:55 2026 -1000"
      },
      "message": "Harden the sample\u0027s JSON dispatcher for embedders\n\nfinbench_dispatch_json_obj() now logs a bounded, control-character-free\nexcerpt of the request instead of the raw body (a caller could forge log\nlines with CR/LF), refuses an \"action\" or \"operation\" that is present but\nnot a string instead of falling through to shape inference, and treats\nexpected_returns and percentiles as Monte Carlo markers. The one shape\nthe fallback cannot resolve, weights and covariance_matrix with nothing\nelse, is documented as routing to portfolioVariance. finbench_log_excerpt\nis exported for adapters.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "8a78396eb4568b26e10319589aa84d18da05108f",
      "tree": "775bac4ff4fa24de0abad29a1032eef7ffb38e8c",
      "parents": [
        "f3f88d32c0719b0b154d4ee37f801523543d6968"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:15:44 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 10:15:44 2026 -1000"
      },
      "message": "Make the URL operation authoritative on the Android static path\n\nA request body may name an operation (\"action\" or \"operation\"), and the\ncamera and audio services dispatch on it, but any authorization in front\nof the server judged the URL, not the body. The engine now reconciles the\ntwo: a body that names a different operation is refused unless the URL\nresolved to a catch-all RESTLocation (\"/\" or none), where the body is the\nonly source by design. Same-operation and unnamed bodies are unchanged.\nHTTP2_ANDROID.md documents the four cases.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "f3f88d32c0719b0b154d4ee37f801523543d6968",
      "tree": "d22982171535ba6abebabb5ffc402515e2e9ec7d",
      "parents": [
        "25bc95a6e7f3852fbc1e20b12977a559d84de03c"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 05:38:21 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 05:38:21 2026 -1000"
      },
      "message": "Budget the correlated book in asset-steps and fix the overflow sentinel\n\nThe correlated monteCarlo budget was n_simulations * n_periods * n_assets^2,\nwhich refused 20 assets at the defaults while the docs promised 100. It is\nnow asset-steps, n_simulations * n_periods * n_assets, which admits 100 at\nthe defaults and about 39 at 100,000 paths; the refusal states the count\nthe budget allows. The terminal value for an overflowing path is 1e150\ninstead of 1e308, which overflowed when squared for the variance or when\ntwo such paths were summed. services.xml\u0027s monteCarlo schema now lists the\nsame properties as the MCP catalog.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "25bc95a6e7f3852fbc1e20b12977a559d84de03c",
      "tree": "decf181aadbf80e3d1107e88b26868d03106703b",
      "parents": [
        "1527ac3cbdc8bcee6080e9d3deccf08a6ff56194"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 05:07:07 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 05:07:07 2026 -1000"
      },
      "message": "Fix compose failure paths, cap its size, and reorder the dispatcher\n\ncomposeCovariance now drops every output array on any failure after\nallocation, so a FAILED reply never carries an unfilled buffer, and is\ncapped at 500 assets: the Cholesky check is O(n^3) and the uniform form\nturns a tiny request into two n x n matrices. The exported json_object\ndispatcher tests the Monte Carlo parameters before the weights rule, as\nthe Android doc requires. Two numeric-edge-case comments now describe\nthe code: paths end as terminal on exp() overflow, and the two-pass\nvariance needs no clamp.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "1527ac3cbdc8bcee6080e9d3deccf08a6ff56194",
      "tree": "c243000c789e64c138c1cecbb244dcc95c64a0ad",
      "parents": [
        "d58b8c9f400bf8fbd42a604f314b6676bebc3c07"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 04:36:57 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 20 04:36:57 2026 -1000"
      },
      "message": "Share the Monte Carlo statistics tail and fix three review findings\n\nThe ~100-line estimator block (two-pass variance, sort, VaR/CVaR,\npercentiles, median) was duplicated between the scalar and correlated\npaths; it is one static helper now, and outputs are bit-identical to the\nduplicated version. A covariance_matrix that is not a JSON array is refused\ninstead of silently running the scalar path. Fixes a signed/unsigned\ncompare in the composeCovariance parser, drops an unused accumulator, and\nexports the json_object dispatcher rather than leaving it unused.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "d58b8c9f400bf8fbd42a604f314b6676bebc3c07",
      "tree": "6eaa8e7612a68a231bc09f88cd1a981b6b07daf8",
      "parents": [
        "beda0b7455b21f9df6520696c06326ebad4a8c11"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 19 15:10:27 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 19 15:10:27 2026 -1000"
      },
      "message": "Add a correlated multi-asset book to the sample\u0027s monteCarlo\n\nWith covariance_matrix and weights present the book is simulated\nbuy-and-hold: per-step shocks are L*Z with L the Cholesky factor of the\nmatrix, each asset on its own GBM, a systemic jump under Merton, and risk\nmeasured on the book value. The response adds simulation_mode, n_assets,\nthe weights used and portfolio_volatility \u003d sqrt(w\u0027Sigma w). One asset\nwith Sigma\u003d[sigma^2] reproduces the scalar run bit-for-bit for the same\nseed on the same machine; the scalar loop itself is untouched.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "beda0b7455b21f9df6520696c06326ebad4a8c11",
      "tree": "5e126abd9d56d7f7e7bc8bdd0c744692c3b3527f",
      "parents": [
        "bc641a62361e1a1fa30c90c114aaf960e74cba34"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 19 15:10:27 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 19 15:10:27 2026 -1000"
      },
      "message": "Pass the operation name to statically linked Android services\n\nThe two-argument Android entry point has no message context, so a service\ncould only infer the operation from the request\u0027s field names, which fails\nonce two operations share a field. try_android_static_service() now adds\n\"operation\": \"\u003cname\u003e\" to the request, the name the URL resolved to, unless\nthe caller already set \"action\" or \"operation\". Documented in\nHTTP2_ANDROID.md; the camera and audio services do not read the field.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "bc641a62361e1a1fa30c90c114aaf960e74cba34",
      "tree": "dc27109affed2460a0e729123f9d83e9d21ea3e7",
      "parents": [
        "a85f9c872f282d0333a1829ce0faf2668b4470d0"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 19 14:33:33 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sat Sep 19 14:33:33 2026 -1000"
      },
      "message": "Add composeCovariance to the financial benchmark sample\n\nBuilds Sigma \u003d D*R*D from per-asset volatilities and either a full\ncorrelation matrix or one uniform rho, then Cholesky-checks it and refuses\na matrix that is not positive definite, naming the failing index. This is\nthe validated front door to portfolioVariance, which by design accepts any\nmatrix. Wired through the handler, the internal dispatcher, the MCP tool\ncatalog, services.xml, metadata and docs. The fuzz harness gains an\nexercise for it and its stale function names now match the header.\n\nCo-Authored-By: Claude Fable 5.1 \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "a85f9c872f282d0333a1829ce0faf2668b4470d0",
      "tree": "67395f057891a3257a84886bf48e2374e7027c4b",
      "parents": [
        "91cad92f1333b9169f7ba7c037668efeae4bd2c6"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Sep 07 08:49:16 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Sep 07 08:49:16 2026 -1000"
      },
      "message": "Bound Monte Carlo periods and drop the redundant covariance zero-fill\n\nCap n_periods independently of the work budget, so a small n_simulations with\na huge n_periods can no longer accumulate the GBM value past DBL_MAX (Inf -\u003e\nNaN response). Make an extreme path terminal rather than capping and then\nmultiplying again on the next period. Remove the covariance zero-fill: the\nshape validation already rejects any matrix that would leave a cell unwritten.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "91cad92f1333b9169f7ba7c037668efeae4bd2c6",
      "tree": "3428c2e04f41333264e581f53185f1a4b46e7d15",
      "parents": [
        "45c56dde23dfa3b1f71095a5e9f3315ae9522bed"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Sep 07 04:37:39 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Sep 07 04:37:39 2026 -1000"
      },
      "message": "Correct the MCP n_assets maximum in the tool description\n\nThe catalog described the limit as 2000; the code allows FINBENCH_MAX_ASSETS\n(3000). Make the description match the constant.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "45c56dde23dfa3b1f71095a5e9f3315ae9522bed",
      "tree": "1e2cd0815883e03b66a275832024e1d6993378c4",
      "parents": [
        "1aa0a15c2736a913fa49b077c30fb4643cf3d906"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Sep 07 04:37:39 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Sep 07 04:37:39 2026 -1000"
      },
      "message": "Harden portfolioVariance and Monte Carlo input validation\n\nZero the covariance matrix before parsing and validate its exact shape, so a\nragged 2D matrix whose element count happens to equal n_assets^2 can no longer\nleave interior cells unread as stale heap. Bound total Monte Carlo work\n(n_simulations * n_periods) so an unbounded n_periods cannot hang a worker.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "1aa0a15c2736a913fa49b077c30fb4643cf3d906",
      "tree": "5d8cc0e0ace45a4068cf4d2bf220f8217002d2f5",
      "parents": [
        "605ffcfa01153ab18583fdf8b418cf6f275eb0ce"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 06 07:02:36 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Sep 06 07:02:36 2026 -1000"
      },
      "message": "doc tweaks on the JSON example code\n"
    },
    {
      "commit": "605ffcfa01153ab18583fdf8b418cf6f275eb0ce",
      "tree": "2da4ec1ac607e2407ccaa4c8fdaf18cb3bf380f8",
      "parents": [
        "37c53ede9174543e50e6507e620202ff453f2feb"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 11:45:09 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 11:45:09 2026 -1000"
      },
      "message": "Remove the SOAP skeleton from the testws sample\n\ntestws_service_handler.c was in no Makefile.am and had not been compiled since\nthe service went JSON-only. It carried a second definition of\ntestws_service_invoke_json alongside axiom-based skeleton entry points, so\nediting it changed nothing. Same case as the login sample\u0027s, removed earlier.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "37c53ede9174543e50e6507e620202ff453f2feb",
      "tree": "57b8e45e9151a4d7c4e434ef903bd9d2106998cf",
      "parents": [
        "054a2b11b2adc4a8042afb2e7d91dca01c0f3851"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 11:39:57 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 11:39:57 2026 -1000"
      },
      "message": "Write down the things the compiler will not tell you\n\nOwnership rules that changed under maintainers (a hash now copies the keys you\ngive it; a property\u0027s free_func can outlive a dlclosed library), the mod_axis2\nallocator whose free has no body, why a sub-package needs its own CFLAGS, and\nthe several ways make check reports green over nothing. Each cost someone real\ntime. Linked from the Developer Guide nav.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "054a2b11b2adc4a8042afb2e7d91dca01c0f3851",
      "tree": "275acc822f32e818253face5fe26e0e271bde6ec",
      "parents": [
        "9ff9fb68edd41f039e23bb3a3b8eb2a770560aa8"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 11:16:05 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 11:16:05 2026 -1000"
      },
      "message": "Remove the empty h2/util placeholder library\n\nThe directory held a Makefile.am and nothing else: libaxis2_h2_util_la_SOURCES\nwas empty, so it produced an 8-byte archive with no members that\nlibaxis2_h2_transport linked for no effect. Drops it from SUBDIRS, LIBADD and\nAC_CONFIG_FILES. Verified both transports still build and test: 44/44 HTTP/1.1,\n39/39 HTTP/2.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "9ff9fb68edd41f039e23bb3a3b8eb2a770560aa8",
      "tree": "a296c36cd8ee358c22f498a5014fdfdb3e6707e7",
      "parents": [
        "0f7f27bdee7a0fd6de6bb25a400b7c11f0613ed7"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 09:15:39 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 09:15:39 2026 -1000"
      },
      "message": "Warn that the two _invoke_json forms are one symbol\n\nThe doc described both signatures without saying they share a name, so a\nbinary cannot hold both and an Android link has to exclude the object carrying\nthe four-argument one. That object usually also carries axis2_get_instance,\nwhich the class-loader path needs, so excluding it leaves the static registry\nas the only route in -- and a service missing from the registry logs at INFO\nand falls through rather than failing.\n\nAdds that, the logcat lines that tell the cases apart, and how to read the\nregistry out of a built binary: --export-dynamic keeps .dynsym through\nstripping, so nm -D plus a short disassembly of android_static_service_lookup\nshows which services are really registered. Registry entries have been lost to\na clean rebuild before; this is how to notice.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "0f7f27bdee7a0fd6de6bb25a400b7c11f0613ed7",
      "tree": "d1c9c395f4bad69ab9af9f75c1ef310d4b7837b5",
      "parents": [
        "ac1ca1de112a484b18dbf60c64acab472cf4b029"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 09:09:01 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 09:09:01 2026 -1000"
      },
      "message": "Register FinancialBenchmarkService in the Android static registry\n\nThe registry knew CameraControlService and AudioSearchService only, so a\nstatically linked build of the financial-benchmark sample had no way in. The\nclass-loader path cannot serve it either: axis2_get_instance lives in the\nhandler object, and that object exports a four-argument _invoke_json that\ncollides with the two-argument form the registry calls, so an Android link has\nto leave it out. Without a weak stub and a lookup entry the adapter an\napplication provides is simply never called.\n\nAdds both, in the shape the file already documents. Android-only; the block is\ninside #ifdef __ANDROID__ and a native build is unchanged.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "ac1ca1de112a484b18dbf60c64acab472cf4b029",
      "tree": "43379dadb2b94a8f91d071b261b184796de22de4",
      "parents": [
        "5bf29378c8e64b78fa3e488ba80bfbe35da0aedb"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 07:23:56 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 07:23:56 2026 -1000"
      },
      "message": "Remove the SOAP skeleton from the login sample\n\nlogin_service_handler.c was in no Makefile.am and had not been compiled since\nthe service went JSON-only. It carried a second definition of\nlogin_service_invoke_json plus axiom-based skeleton entry points, so editing it\nchanged nothing and reading it suggested a SOAP path that no longer exists.\n\nThe HTTP/2 userguide listed it, and testws_service_handler.c, as required and\ncomplete; both claims were wrong, and its file-existence check tested for names\nthat are not in either Makefile.am. Corrected to the files actually built.\ntestws_service_handler.c is the same kind of leftover and is noted there rather\nthan removed.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "5bf29378c8e64b78fa3e488ba80bfbe35da0aedb",
      "tree": "436d32ae68f3c1919c362e566da06fb04570dad4",
      "parents": [
        "3189289769607bdb9b186c3830db8e7b1fa73f91"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 07:09:22 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 07:09:22 2026 -1000"
      },
      "message": "Say what the login sample actually does\n\nIt builds again on GCC 15: at_pos took strchr\u0027s result off a const string into\na non-const pointer, which -Wdiscarded-qualifiers flags and samples\u0027 -Werror\nmakes fatal. It is only read, so it is const now.\n\nThe README described three operations -- authenticate, validateToken, logout --\nthat do not exist, and services.xml declares the same three. The only\nimplemented operation is doLogin, and that one is not declared. The README now\nsays so, along with the fact that the token is unsigned despite an HS256\nheader, that validate_jwt_token only counts full stops, and that\nlogin_service_handler.c is never compiled. The Java guide this was ported from\nputs JWT and JWE out of scope, so the shallow token is faithful to it; the\ndocumentation claiming otherwise was not.\n\nNo behaviour change: the descriptor and the unused code are left alone.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "3189289769607bdb9b186c3830db8e7b1fa73f91",
      "tree": "215536ff0d128d85d9a2cbd1719b7790c61ffa43",
      "parents": [
        "221cbf52f6c32ad9d71ab9f82a91c6a658316277"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 06:32:01 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 06:32:01 2026 -1000"
      },
      "message": "Stop INSTALL requiring libgtest-dev\n\nEvery test Makefile.am names the archive the build itself compiles, so no\nlibrary has to be findable on the linker path and the LDFLAGS workaround this\ndescribed is no longer needed. f7319af01 made that change; INSTALL was not\nupdated with it. Confirmed on a box with no Google Test package installed.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "221cbf52f6c32ad9d71ab9f82a91c6a658316277",
      "tree": "01413abe999407183f5ca604f3cdf82147388406",
      "parents": [
        "8ff3ed4ab2c808419f9e75bf42ef9ec26a89a776"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 04:20:17 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 04:20:17 2026 -1000"
      },
      "message": "Run the HTTP/2 transport in CI as well\n\nThe two transport configurations build disjoint test sets, and only the\nHTTP/1.1 one had ever run: h2_json_integration_test,\nh2_performance_benchmark_test and test_http_service_provider_interface had\nno CI coverage, which is the path the JSON services actually serve. Adds a\nmatrix leg for --enable-http2 and libnghttp2-dev to go with it.\n\nThe two deployment tests dlopen the HTTP/1.1 sender and receiver, which that\nconfiguration does not build, so they now skip rather than fail. The sender\none never asserted on the result at all -- it reported success with no\nlibrary present.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "8ff3ed4ab2c808419f9e75bf42ef9ec26a89a776",
      "tree": "a3b7bcfe01643b8f39af2ac3d5f8eb5eac7f6206",
      "parents": [
        "21c8b547c9ce1e8d5bd63593c45e72627065ee3c"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 04:20:17 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Sep 04 04:20:17 2026 -1000"
      },
      "message": "Stop the payload generator\u0027s size guards underflowing\n\nstrlen returns size_t and target_size_bytes is int, so target_size_bytes -\n2000 was evaluated in int, went negative for any target under the margin and\nthen converted to a huge size_t -- the guard never fired and the generator ran\npast its buffer. Comparing signed also gives the right answer: a target\nsmaller than the margin stops at once. Same shape as the url.c defect.\n\nFound by -Wsign-compare on GCC 11, which is stricter here than the local\ncompiler; the flag only started reaching this file this week.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "21c8b547c9ce1e8d5bd63593c45e72627065ee3c",
      "tree": "b93d8a27b38ab42a84183c8516563decece5e828",
      "parents": [
        "630fe6004f9c7d505cedab2b910744b012c40bac"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 06:59:37 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 06:59:37 2026 -1000"
      },
      "message": "Remove the Visual Studio and win32 build files\n\nides/vc, ides/vc6, ides/vc9 and build/win32 -- 133 files. The project files\nare .dsp and .vcproj, formats no Visual Studio since 2010 can open, and the\nlast substantive change to any of them was 2011. They still carried build\nrules for TCP, CGI, NTLM, IIS and tcpmon, all removed earlier, and\ndist_with_ntlm depended on a target that never existed under that name.\nINSTALL already called Win32 legacy and unsupported and pointed Windows users\nat WSL2, which is now the only documented route.\n\nThe dist-hook line stripping /WX went too: build/win32/makefile was the only\nfile it matched, and with it gone the sed ran with no input and failed dist.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "630fe6004f9c7d505cedab2b910744b012c40bac",
      "tree": "9fe593f8b6ed623e56b005b502f684f85b292330",
      "parents": [
        "6ca9a4818d3d4359ac055b1995fcdde0ceabae4f"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 05:27:05 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 05:27:05 2026 -1000"
      },
      "message": "Remove the CGI remnants left behind in January\n\nAXIS2C-1708 deleted the CGI sources but left the win32 nmake target that\nbuilds them, an INSTALL section telling users to set up cgi-bin, and two doc\nreferences. The win32 dist target depended on a rule whose sources no longer\nexist. Also drops the Raw TCP Transport section from the HTTP/1.1 doc, stale\nfor the same reason. ChangeLog, NEWS and the release notes keep their CGI\nentries: those are the record of the removal, not leftovers.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "6ca9a4818d3d4359ac055b1995fcdde0ceabae4f",
      "tree": "ecf623f1ff194f05dd84673cb74ec0b6644126f3",
      "parents": [
        "07e1bc2285c07b672f8b834d2e9d0b1b321969fe"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 05:08:45 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 05:08:45 2026 -1000"
      },
      "message": "Remove tcpmon\n\nA plain byte relay with no TLS, HTTP/2 or HPACK support, so it cannot read\nthe traffic this project now serves: JSON requires HTTP/2, which requires TLS\nALPN. Its own code had not changed functionally since 2013, and it had not\nbeen built since January, when AXIS2C-1708 dropped it from SUBDIRS while\nstating it was retained. Wireshark, mitmproxy and mod_dumpio all do the job\nproperly now.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "07e1bc2285c07b672f8b834d2e9d0b1b321969fe",
      "tree": "49e0f2b5ef5b364d78b4a6102a51a2e3b11b56b6",
      "parents": [
        "f8e0b9b7fbf3cf17adadc2dff99adc9e51398876"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 04:00:34 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Sep 03 04:00:34 2026 -1000"
      },
      "message": "Set the char signedness flag in the remaining sub-packages\n\nsamples, tools/md5 and tools/tcpmon have their own GCC flag blocks and were\nmissed by the previous commit. samples builds with -Werror, so it gets\n-fsigned-char only -- a signedness warning there would fail the build rather\nthan inform anyone. tools/tcpmon is not built at all; the edit is for\nconsistency if it is ever revived.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "f8e0b9b7fbf3cf17adadc2dff99adc9e51398876",
      "tree": "55cd30ff844ead42025a2e5825c5b14c084a00d4",
      "parents": [
        "b439253fd83c726c44d5c7a545be7746e492f437"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 18:40:19 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 18:40:19 2026 -1000"
      },
      "message": "Pin char signedness, and set the flags in every configure.ac\n\naxis2_char_t is plain char: signed on x86, unsigned on ARM. The code assumes\nthe signed reading -- ctype calls pass one without the (unsigned char) cast --\nso the arm64 build was a different program from the x86 build it is tested\nagainst. The sub-packages are AC_CONFIG_SUBDIRS children and do not inherit\nCFLAGS from the root, so -Wformat-security and -Wsign-compare had been reaching\nsrc/ only; all three flags are now set in all five.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "b439253fd83c726c44d5c7a545be7746e492f437",
      "tree": "75c576dedfd42a754fd1c1cd1c127244852b1b8f",
      "parents": [
        "c79e84cd89df619947ae21629b3c4a3a16e61310"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 11:11:28 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 11:11:28 2026 -1000"
      },
      "message": "Say so when make check runs no tests at all\n\nEvery test Makefile.am gates TESTS on GTEST_AVAILABLE, so --enable-tests\nwithout --with-gtest compiles out all 38 suites and automake still prints\n\"All tests passed!\" and exits 0. Warn at configure time and print a banner\nfrom check-local, which runs after the recursion because \".\" is not in\nSUBDIRS. Silent when gtest is present, and when tests were never requested.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "c79e84cd89df619947ae21629b3c4a3a16e61310",
      "tree": "a44a09788230fa1c9a361af459cd4f311a3947b8",
      "parents": [
        "f7319af01e861882fd1947e4d7f01bfb6dc09a61"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 10:09:56 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 10:09:56 2026 -1000"
      },
      "message": "Pass both arguments to the two-conversion ADB name patterns\n\nThe guard looked for \"%s_%s\", which adb_%s_get_%s and adb_%s_set_%s do not\ncontain, so both fell to the single-argument branch and the trailing %s read\nan unset register. Count the conversions instead. ASan in CI caught this as\na SEGV in snprintf; it survives locally only because the garbage pointer\nhappens to be readable.\n\nCo-Authored-By: Claude \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "f7319af01e861882fd1947e4d7f01bfb6dc09a61",
      "tree": "a54ff816cabeeba4b999df0c792fb015677e33be",
      "parents": [
        "4b2aa7e5cee8cf714724871c2afdf116409adb13"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 09:16:14 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 09:16:14 2026 -1000"
      },
      "message": "Link the util tests against the gtest built from source\n\nEleven of them pulled in -lgtest and -lgtest_main, the distro binaries,\nwhile the rest of the tree and util/test/allocator use the copy\nutil/gtest builds from the --with-gtest sources. That only worked where\nlibgtest-dev happened to be installed; the CI runner has the sources\nand not the binaries, so the build failed at link. They now match\nallocator, guard included, so --enable-tests without --with-gtest skips\nthem instead of failing.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "4b2aa7e5cee8cf714724871c2afdf116409adb13",
      "tree": "de3052396ddffd09f509018847442a468f280edb",
      "parents": [
        "dc3cd2f2a297b62ab7f9215d698e1d8cb5ddfc08"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 07:36:50 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 07:36:50 2026 -1000"
      },
      "message": "Keep transport libraries mapped for as long as their pointers live\n\nA transport hands out function pointers that outlive it. The HTTP\nsender puts the http client into a message context property whose free\nfunction lives in libaxis2_http_sender, and on the client that property\nsits in the map shared with axis2_options, released by\naxis2_options_free -- which svc_client_free runs after\naxis2_conf_ctx_free has unloaded that library. Every sample client\njumped into unmapped memory at exit. The transports\u0027 descriptors now\ndetach the library instead of closing it, which cost two things: the\nhandle setter rejected the NULL that detaching needs, and freed the old\nhandle with AXIS2_FREE though dlopen never allocated it.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "dc3cd2f2a297b62ab7f9215d698e1d8cb5ddfc08",
      "tree": "8ad5ac1063a15ec7b54c477afbda9aee77396f91",
      "parents": [
        "81f034732a0ebddf501b104574f9ad0826874bed"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 06:11:13 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Sep 02 06:11:13 2026 -1000"
      },
      "message": "Run the test suite under AddressSanitizer in CI\n\n--enable-asan has been there all along with nothing running it, and the\ndefects it catches are the ones review does not: this month it found a\nuse-after-free on the client response stream that reading the same code\nhad missed. The job configures with --with-gtest, because without it\nmake check builds no gtest suites and reports a green run of zero\ntests, and points AXIS2C_HOME and LD_LIBRARY_PATH at the tree under\ntest so another install cannot be tested instead. Leak detection stays\noff until the known leaks are cleared.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "81f034732a0ebddf501b104574f9ad0826874bed",
      "tree": "04ec96b47d72eb3c269bf95a891fdf1595eb03bb",
      "parents": [
        "7a331a71b95799ee858c903fb6b3b6071aebf32f"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 01 10:54:20 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 01 10:54:20 2026 -1000"
      },
      "message": "Say that make install replaces axis2.xml\n\n$(prefix)/axis2.xml is a symlink into share/axis2c, and that file is\nreinstalled every time, so an enabled transport or an attachmentDIR set\nbefore installing is silently gone afterwards. Nothing said so at the\npoint it bites. The install now says it, and the testing guide gives\nthe ordering -- configure after the last install, samples build\nincluded, since that installs again.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "7a331a71b95799ee858c903fb6b3b6071aebf32f",
      "tree": "e8ec57b08ece653d241e6db6ecbeff6991c7a035",
      "parents": [
        "282417f59d5dd0ba1364dae4bc0b01a117c6c634"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 01 10:54:20 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 01 10:54:20 2026 -1000"
      },
      "message": "Give the response body stream a single owner on the client\n\nThe stream came from axis2_http_simple_response_get_body, which the\nresponse frees when stream_owned is set, and was then handed to a\nmessage context property carrying a stream free function. Both claimed\nit. A later request replaces the AXIS2_HTTP_CLIENT property, that free\nreaches the stream through the response, and replacing\nAXIS2_TRANSPORT_IN then frees it again -- a use-after-free ASan reports\nin axutil_stream_free. The property keeps it; the response lets go.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "282417f59d5dd0ba1364dae4bc0b01a117c6c634",
      "tree": "e2b2cf2f71fc98340eea38bdd185e05a37d3517a",
      "parents": [
        "0ba2ad02f50abba9ea9bc40756e797d06a5a7ada"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 01 05:07:27 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Sep 01 05:07:27 2026 -1000"
      },
      "message": "Stop freeing hash keys the hash already owns\n\nSince AXIS2C-1632 a hash_set with AXIS2_HASH_KEY_STRING copies the key,\nand axutil_hash_free releases the copy -- as does deleting an entry.\nFour teardowns still walked their entries freeing keys first, which\nfrees each one twice. The MTOM path aborted the standalone server on\nany request that reached soap_builder teardown; the others are the same\nmistake waiting on their own paths. Values stay the callers\u0027 business.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "0ba2ad02f50abba9ea9bc40756e797d06a5a7ada",
      "tree": "d7c2efc7aafb8fa2b8e74a05fd897c86ebcff8cb",
      "parents": [
        "bac89c033fb6cc031d71706e7693ab3dc3bec331"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Aug 31 07:54:53 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Mon Aug 31 07:54:53 2026 -1000"
      },
      "message": "Build the standalone HTTP server again\n\naxis2_http_server has not been produced by any configuration since\nsimple_axis2_server left SUBDIRS when the HTTP/2 transport landed. The\nremoval suits HTTP/2 JSON mode, where httpd carries the transport, but\nit was unconditional, so the HTTP/1.1 SOAP build lost the binary its own\nquickstart tells people to run. Nothing complained because configure\nstill generates the Makefile. Gated on !WITH_NGHTTP2 like the HTTP/1.1\ntests, with DIST_SUBDIRS naming it so make dist still recurses.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "bac89c033fb6cc031d71706e7693ab3dc3bec331",
      "tree": "1e1ec4852590ccccf5b6dfed2b4d9816e233ac73",
      "parents": [
        "76da49e86dedfeb0ca3d6b30ef77961693a65158"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Aug 30 11:17:53 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Aug 30 11:17:53 2026 -1000"
      },
      "message": "Cover attachment caching, and stop gitignore hiding test sources\n\nFive tests over the attachmentDIR path: that a large attachment reaches\na file, that its name carries nothing of the sender\u0027s Content-ID, that\ntwo requests sharing a Content-ID get separate files, and that a file\nalready in the directory is left alone. Against the previous naming\nthree of them fail. The gitignore change is what let the source be\ncommitted at all -- **/test_* matches a new test\u0027s source as readily as\nits binary, so test_mime_parser.cc was silently unaddable.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "76da49e86dedfeb0ca3d6b30ef77961693a65158",
      "tree": "110dfeff0b4f2d2e000bd4e9fbbeb91c1bc755e8",
      "parents": [
        "8a075d7f5be49a93851548669c400ba1756e764b"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Aug 30 04:48:13 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Sun Aug 30 04:48:13 2026 -1000"
      },
      "message": "Reuse an already-declared namespace instead of leaking a second one\n\naxiom_element_declare_namespace returns SUCCESS without taking the\nnamespace when the element already declares that prefix and uri, so the\nbuilder\u0027s object was left owned by nobody -- declared_namespaces is a\nlookup cache and axutil_hash_free does not free values. The empty-prefix\nlookup matches on uri alone, so \"xmlns:p\" followed by a default xmlns\nfor the same uri is enough. Ask before creating rather than freeing\nafter: the builder reads the namespace on the next line and caches it.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "8a075d7f5be49a93851548669c400ba1756e764b",
      "tree": "bf4f5bd1fededb637d38dd7018739df328d63543",
      "parents": [
        "242cd75ce8892065e9ca5c27416c0ed4c22b09eb"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 12:20:43 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 12:20:43 2026 -1000"
      },
      "message": "Generate MTOM cache file names instead of taking them from the sender\n\nThe cache path was attachmentDir plus the URL-encoded Content-ID, so\nthe sender chose which file was opened: two requests carrying the same\nContent-ID landed in one file and each other\u0027s attachment bytes, and a\nContent-ID matching something already in the directory opened that\ninstead. The name is now generated per attachment, and the parser keeps\na Content-ID to path map because the retrieval side can no longer\nrebuild the name.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "242cd75ce8892065e9ca5c27416c0ed4c22b09eb",
      "tree": "90576bb7493b8ea87e35b5ac6d78441b68d23770",
      "parents": [
        "727677f2de45483db4fe723eb398e5607004fcae"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 11:46:45 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 11:46:45 2026 -1000"
      },
      "message": "Copy the strings and nodes the policy model outlives\n\nThe rp model stored axiom attribute values, element text and nodes\nstraight from the document it was parsed from, and the free functions\nleft them alone on the grounds that they were borrowed. Nothing kept\nthe lender alive: a deploy-time policy is built from the services.xml\ntree and held in policy_include, and the description builder frees that\nwhole tree when deployment finishes, so every later read is of freed\nmemory. Setters now copy, free functions release, and the two node\nfields clone as neethi_assertion_set_node already did.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "727677f2de45483db4fe723eb398e5607004fcae",
      "tree": "9f062ae8f0398de44ffa4b7bac5c28fec30890d5",
      "parents": [
        "e379cb7bf80dfdb655525375bfdcb176e51a7dd3"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 04:05:28 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 04:05:28 2026 -1000"
      },
      "message": "Remove the unused prepend stream free function\n\nIt was superseded by the managed-stream convention the create function\nalready follows: buffer_head holds the prepend buffer and fp the\nunderlying stream, and axutil_stream_free\u0027s MANAGED case frees both\nplus the wrapper. This function frees the same three allocations\nthrough the extended fields that alias those handles, so wiring it up\nwould have double-freed all of them. A note at the site records why\nthere is no free function rather than leaving that to be rediscovered.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "e379cb7bf80dfdb655525375bfdcb176e51a7dd3",
      "tree": "bc445bcfc57d8600db12d10d1ad847d7ffc697c0",
      "parents": [
        "532bd75dcc7a752cb14595a1bc3cefc204de8f04"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 03:40:11 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Fri Aug 28 03:40:11 2026 -1000"
      },
      "message": "Write the whole chunk the chunk header announced\n\naxutil_http_chunked_stream_write emitted a header naming count bytes\nand then wrote whatever one stream write accepted, which for a socket\nis whatever fits in its send buffer. Callers cannot recover from a\nshort return there: two of the three respond by calling again with the\nentire buffer, so a second chunk header lands inside a chunk that was\nnever finished, and it repeats while the writes keep coming up short.\nComplete the announced body or fail.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "532bd75dcc7a752cb14595a1bc3cefc204de8f04",
      "tree": "a1a9722921513187856f94d92813db05e3802cf7",
      "parents": [
        "d73bebe2b37e8320b67d9fb8903682488cafbfe1"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Aug 27 14:33:12 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Aug 27 14:33:12 2026 -1000"
      },
      "message": "Turn on -Wsign-compare so the signed length stays readable\n\nA signed length compared against a size_t converts to a large positive\nvalue, which is how a -1 error return reads as \"long enough\" and slips\npast a bounds test. That is the half of the axis2_ssize_t change the\ntype alone cannot fix, and the warning is in -Wextra rather than -Wall\nfor C, so it was off. Two byte-count loops in http_client mixed int\nagainst size_t; with those in agreement the tree compiles clean under\nit.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "d73bebe2b37e8320b67d9fb8903682488cafbfe1",
      "tree": "b2d6331d8a5e9755400ffd109500f371e7f280a0",
      "parents": [
        "2ddf163507879af8e30e7ec5c0b9774b816927d6"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Aug 27 08:47:12 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Thu Aug 27 08:47:12 2026 -1000"
      },
      "message": "Skip the standalone-server tests when no receiver is configured\n\nThe shipped axis2.xml comments out the http transportReceiver and\ntransportSender because HTTP/2 JSON mode serves through httpd, so the\ntwo tests that need a live standalone server could never pass in the\ndefault configuration. They failed instead of skipping, and a failing\ntest stops make check recursing: the suite reported 19 tests where 43\nexist. The helper now says whether the server came up, and both tests\nskip with the configuration to change.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "2ddf163507879af8e30e7ec5c0b9774b816927d6",
      "tree": "2034ad5a08c8068d0582c11a8496ef86aa837599",
      "parents": [
        "bbdca6f35513a713b6180c275d73965eb163f799"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Aug 26 06:44:09 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Aug 26 06:44:09 2026 -1000"
      },
      "message": "Document why a stale install silently replaces the tree under test\n\nLD_LIBRARY_PATH is searched before RUNPATH, so an Axis2/C install left\non that path is loaded ahead of the freshly built tree. From a\nbuild_for_tests.sh install it is ASan-instrumented and the test aborts\nbefore main with a message that names neither cause; from any other\ninstall the tests simply run against code nobody just compiled. Both\nturn \"what is installed on this box\" into test results.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "bbdca6f35513a713b6180c275d73965eb163f799",
      "tree": "9fb191629628c9398e58c84dcb0272e15eebe4cf",
      "parents": [
        "6d9c524dadbb3fdda6571c7490ab5242b42a4152"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Aug 26 06:43:16 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Aug 26 06:43:16 2026 -1000"
      },
      "message": "Stop the mock HTTP server resetting the connection mid-response\n\nsecurity_many_headers was not a platform failure: it fails about four\nruns in ten on one machine. The helper closed the client socket with\nthe request\u0027s trailing bytes still unread, which sends RST rather than\nFIN and kills whatever the client had not yet read -- visible only once\na response is large enough that the client is still reading, hence the\n10,000-header case and nothing else. Half-close and drain first. The\nsingle-shot send() also ignored its return value, which truncates any\nresponse past the socket buffer.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "6d9c524dadbb3fdda6571c7490ab5242b42a4152",
      "tree": "5fb119ad090a6562927255d68fcaf393c4aefe8c",
      "parents": [
        "a4f9bbd281ee197c8bccf05aba63d2909f27bdda"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 08:51:35 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 08:51:35 2026 -1000"
      },
      "message": "Free each mod_axis2 JSON stream once and own only allocations\n\nThe request and response streams were freed on the way into the engine\nblock and again on every path out of it, the IS_JSON_STREAM property\nowned the constant 1 and sent it to free at teardown, and the fault\npath read a stream that freeing fault_ctx had already released. None of\nthis bites today: mod_axis2 installs an allocator whose free is a no-op\nfor per-request pool memory, so the frees do nothing and the pool\noutlives the read. Correct the ownership rather than leave it resting\non that.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "a4f9bbd281ee197c8bccf05aba63d2909f27bdda",
      "tree": "11286fe74108c38db246800ecfd22b4765e2f980",
      "parents": [
        "776f8d96f504b04a993050675b3a815fe1ca2230"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 08:51:22 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 08:51:22 2026 -1000"
      },
      "message": "Make axis2_ssize_t signed so its error returns exist\n\nIt was unsigned int, so the -1 that axutil_strlen and\naxutil_string_get_length document for failure arrived as UINT_MAX and\nevery \"\u003c 0\" test against one was unreachable. Same width and alignment,\nso the soname stays at 0 and compiled consumers receive identical\nbytes; revision moves in all five configure.ac files. Signedness alone\nis not enough -- an unsigned operand on the other side still converts\n-1 to a large positive value, which is why url.c stops comparing a\nlength against sizeof.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "776f8d96f504b04a993050675b3a815fe1ca2230",
      "tree": "639f43c09467232895b10029ff8c48d91d5e0b8b",
      "parents": [
        "9be8e306a388139eaf8472d8a2e8139ac5fd7a39"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 05:25:04 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 05:25:04 2026 -1000"
      },
      "message": "Remove the unreferenced autotools integration script\n\nNothing invokes it -- not a Makefile, not another script, not CI -- and\nit has not been run since it was added. A test that never runs reports\nnothing while looking like coverage.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "9be8e306a388139eaf8472d8a2e8139ac5fd7a39",
      "tree": "cb68a6a5e20aeb9a8bd85a973f0a030cb1b9435c",
      "parents": [
        "18e0900876b886c3a4de882f5fb4bd0cdc2bb0c5"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 04:41:28 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 04:41:28 2026 -1000"
      },
      "message": "State what holds memory safety, not how risky it feels\n\nEvery row of the memory safety table rated itself Low and every rating\nwas wrong. A rating is not testable and decays as the code moves; the\nmechanism column is what an external review could actually engage with,\nbecause a claim that names a mechanism can be shown false. Ratings are\nreplaced by what each category does not cover, which is where the\nunsigned axis2_ssize_t is now recorded. -Wformat-security has moved\ninto configure, so production builds no longer have to pass it.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "18e0900876b886c3a4de882f5fb4bd0cdc2bb0c5",
      "tree": "2c464017fa02fff6c9964fb8f87245acf7b8ff89",
      "parents": [
        "5121c91577de72b2efbdf237a8908608b0e8d3ca"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:58 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:58 2026 -1000"
      },
      "message": "Bound policy nesting and normalization, guard the accessors\n\nPolicy parse and normalization recurse once per nested element with\nnothing upstream bounding document depth, and the cross product is\nexponential in sibling alternatives; both are capped. The builders read\nthe first alternative\u0027s value unchecked, which is NULL when\nnormalization failed or the list is empty, so the shared accessors now\nanswer defensively instead of faulting.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "5121c91577de72b2efbdf237a8908608b0e8d3ca",
      "tree": "b3908db62a45971e4d80a2de2211f8a92901d71f",
      "parents": [
        "d5d60c9fd106c949c99e5fb17047b1bd00737b5e"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:51 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:51 2026 -1000"
      },
      "message": "Fix XPath expression ownership and token buffer bounds\n\nThe context freed the caller\u0027s compiled expression on attach and again\non free, so re-evaluating one used it after free; it now borrows.\nLiteral and number operators aliased the expression\u0027s parameters into\nresult nodes that evaluation frees. NCName and literal copies had no\nbound on their 255-byte frame buffers.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "d5d60c9fd106c949c99e5fb17047b1bd00737b5e",
      "tree": "eb14d86b4818b3d3bd2eabcb3d9485b065315e94",
      "parents": [
        "d9c70e36064ea9be04caf07220190632238e6afe"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:44 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:44 2026 -1000"
      },
      "message": "Give shared request objects exactly one owner at teardown\n\nThe OUT context aliases the IN context\u0027s transport stream, transport\ninfo and To epr, so freeing it without the reset gives the IN context a\nsecond free of each, and both failure exits left it registered in the\nop_ctx for worker cleanup to free again. The JSON response was owned by\ntwo properties, and set_fault_to wrote the To field.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "d9c70e36064ea9be04caf07220190632238e6afe",
      "tree": "7e10598a13b42cf3d58db8003ebb71a9bdca9d8e",
      "parents": [
        "70a71a0f6c6946354e02ba3d49010230175a4deb"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:36 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:36 2026 -1000"
      },
      "message": "Keep the XML parsers inside their buffers and their stack\n\nAn element whose name did not start with a name character emitted an\nempty-element event without pushing a frame, so the next call popped an\nancestor\u0027s frame and eventually a NULL. next_no_char copied the bytes\nasked for whether or not the refill supplied them. The libxml2 wrapper\nchecked against the sum of two limits, which is neither limit.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "70a71a0f6c6946354e02ba3d49010230175a4deb",
      "tree": "08eb5474e4e2823fe7ba98b48edae7f53d809407",
      "parents": [
        "9fb230f658b6feca814d136d1010b38c6334bce1"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:29 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:29 2026 -1000"
      },
      "message": "Harden the MIME parser against malformed multipart bodies\n\nA failed read left the buffer slot cleared and the boundary test then\ndereferenced it. Header scans stepped past the terminator when the\ndelimiter was absent, and the CRLF deductions ran without checking\nthere were two characters to deduct. The buffer array is now held to\nits max_buffers bound and the part count is capped.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "9fb230f658b6feca814d136d1010b38c6334bce1",
      "tree": "e82daa2678a80fcc7364b86d5a1e7747c978ea2c",
      "parents": [
        "0159bc0404f435962e071125f06aa95a61ba432b"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:19 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:19 2026 -1000"
      },
      "message": "Bound every request body by maxRequestSize\n\nThe ceiling was enforced only against a positive Content-Length and in\nthe chunked branch. A body with content_length -1 skips the early\nreturn by design and unread_len is no bound, so the non-chunked branch\nread until the peer stopped -- the state mod_axis2 leaves every chunked\nrequest in. Both branches now meter, and a POST or PUT with neither\nheader gets 411.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "0159bc0404f435962e071125f06aa95a61ba432b",
      "tree": "92c58ee147625f95642012c5bdeba794a3154fe1",
      "parents": [
        "37fcfebfe5659c7a6b26be1cfde3912eff417a87"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:12 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:12 2026 -1000"
      },
      "message": "Fix JSON flag setup and require --enable-json for HTTP/2\n\nPKG_CHECK_MODULES ran after the appends that use JSON_CFLAGS, so the\njson-c include path never reached CFLAGS and it only built where json-c\nsits on the default include path. The HTTP/2 transport links json-c\ndirectly, so --enable-http2 without --enable-json now fails at\nconfigure rather than at link.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "37fcfebfe5659c7a6b26be1cfde3912eff417a87",
      "tree": "70a9fa95cbbb6d01b040919961e89beb85b4fc3a",
      "parents": [
        "a8b0a38799d98905be1aaf621903a0892f910c0e"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:03 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Tue Aug 25 03:23:03 2026 -1000"
      },
      "message": "Drop docs for the removed libcurl backend and leak\n\nThe libcurl backend went before 2.0.0, but the userguide still\ninstalled libcurl4-openssl-dev and passed --enable-libcurl, which\nconfigure no longer accepts. The known-limitations page described a\nconnection-map leak that AXIS2C-1632 fixed, at a line past the end of\nthe file.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    },
    {
      "commit": "a8b0a38799d98905be1aaf621903a0892f910c0e",
      "tree": "6dfdcd832e3fbed4812c0ee2f4a948e531bb6f90",
      "parents": [
        "d81e2f38583ce0d7623b3fd1e8a9492c2e7f8307"
      ],
      "author": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Aug 19 09:30:18 2026 -1000"
      },
      "committer": {
        "name": "Robert Lazarski",
        "email": "robertlazarski@gmail.com",
        "time": "Wed Aug 19 09:30:18 2026 -1000"
      },
      "message": "Set GitHub topics and repo description via .asf.yaml\n\nThe repo had no topics and a description of \"Apache Axis2-C\", so it does not\nsurface for anyone browsing or searching GitHub for a C framework doing JSON,\nREST, httpd modules, HTTP/2, Android or MCP. Adds 16 labels and a description\nnaming what the framework actually does.\n\nPurely additive; the existing homepage and branch-protection ruleset are\nunchanged. Topics apply once ASF infrastructure processes this on the default\nbranch.\n\nCo-Authored-By: Claude Opus 5 (1M context) \u003cnoreply@anthropic.com\u003e\n"
    }
  ],
  "next": "d81e2f38583ce0d7623b3fd1e8a9492c2e7f8307"
}
