blob: 85c278eefc254ee06092e10fccac05054c4254d1 [file] [log] [blame]
# Nice to have features #
* If Apache::SubRequest::DESTROY gets invoked more than once on the
same object (directly and indirectly via object's exit of scope),
memory gets corrupted and some later unrelated requests get affected
badly (segfaults in unrelated areas). So we probably need to do the
same handling as APR::Pool, where we make sure that no matter how
DESTROY is called, it's always called only once, any consequent
calls can be made a NOP or assert. BTW, it's easy to reproduce the
problem, by simply calling $subrequest->DESTROY 3 times or so in a
raw, the server just hangs...
* 'apachectl configtest' will not test perl modules, unless <perl>
section or PerlLoadModule has happeneded to be loaded (since by
default we start perl in the post-config phase. A possible solution
is to try to detect 'apachectl configtest' and force the interpreter
loading during the config phase, so that configtest will actually
work for mp2 too. there is also an opinion that there is no
problem, and configtest just checks the config file syntax. But the
problem is that users might be mislead and think that it tests
modperl too, and when doing a restart it may fail, even though
configtest said it's OK.
* PerlPreConnectionHandler is implemented, but the 'void *csd' arg in
the callback is ignored. will require a modification of
modperl_callback_run_handlers to pass yet another optional argument.
Status: most likely nobody will ever want to use this option, so
don't waste time working on it.
* Apache::FakeRequest:
since APR can be used outside of httpd, and we can alloc request_rec
and similar structures, it should be possible to bootstrap an
inside-httpd interface and outside-httpd interface. its not really
worthwhile looking at until APR actually installs its *.so or *.a
libraries somewhere apxs can find them. and, there's a bunch of util
functions (e.g. URI stuff) that is supposed to move from httpd into
* could add support for embedding apache directives into perl
directives values, e.g. interpolating $ServerRoot
PerlSwitches -I$ServerRoot/lib/perl
* Since now we have protocol modules, it'd be nice to have a similar
thing to PerlCleanupHandler, which works only for HTTP requests.
It should probably be called PerlConnectionCleanupHandler.
If we add it we should probably rename PerlCleanupHadndler to
PerlRequestCleanupHandler and keep the old name as a deprecated alias.
We could also have PerlServerCleanupHandler, but that's exactly what
PerlChildExitHandler does. Consider having ServerCleanup as an alias.
config features:
- tie %ENV to r->subprocess_env so stores are added to
r->subprocess_env and fetches are looked up in there and elsewhere
(e.g. HTTP_* from r->headers_in). see modperl_env.c, current
implementation is not threadsafe and requires 5.7.2+
- make 'PerlSetVar $Foo value' work like 'local $Foo = value'
for the given location
- allow Perl*Handler's to have arguments in config files
- allow <Perl></Perl> configuration sections to have read access to internal
configuration structures (would be nice if we could tie a %namespace::)
- setuid/gid before running any Perl code
- implement PerlINC (or similar) as a nicer interface for the working
PerlSwitches -Mlib=/home/dev1/lib/perl, to set different @INC for
different virtual hosts.
See the thread:
- a possible implementation of PerlOptions +Inherit, similar to
+Parent but which allows virtual hosts to inherit everything that
was loaded by the main server, at the point of their definition in
the config file. This can make it easier to write configuration
files where there is a common base of modules to be loaded in all
servers. Of course this can be done by putting all these common
modules and code into and running it from the base server and
all virtual hosts.
- PerlModule can be made more efficient using Perl_load_module
Status: it's possible to implement, but currently there is no way to
prevent from Perl logging the loading error messages to the
perl language features:
- @ARGV magic, tie to query string
- sub handler : method ($) {}
call $class->new($r) and pass the returned object as the first and
only arg to the method handler
- some mod_perlIO/PerlIO type methods for xs modules? (e.g. Apache::Peek)
- possible to support BEGIN,CHECK,INIT blocks similar to how END is
supported via ModPerl::Global::special_list_{call,clear}
optimization features:
- copy-on-write SvPVX
- hook Perl malloc into apr_pool
- for "compiled handlers" w/ ithreads manage SV allocation via
server-lifetime apr_pool_t
- "garbage collector" thread to walk padlists looking for certain things
worth releasing.
- "mip manager" thread to watch # of active interpreters,
cloning/destroying when needed (rather than waiting for a request to
- use subpools per-callback/handler (might trim some memory bloat)
note: creating subpools requires a malloc mutex lock with threaded
- improve the "stacked handlers" implementation, including:
+ allow push_handlers to have an additional argument, an array ref,
which will be passed to the handler as arguments, e.g.
$r->push_handlers("PerlHandler", \&some_sub, ['one', 'two', 'etc']);
- might add an alias for $filter->connection (now we have $filter->c), to be
more intuitive since we have $r->connection.
- core Apache::SubProcess w/ proper CORE::GLOBAL::{fork,exec} support
+ currently works only with $] >= 5.007003 (see the
apache/subprocess test)
- It's possible that we will add:
if MP_APACHE_COMPAT Makefile.PL option is true. But this adds bloat,
so this is just an option to consider.
- Apache::File->tmpfile now lives only in compat. Consider adding
APR::File->mktemp (apr_file_mktemp) and a perlio layer
defined in terms of apr_file_t to use it.
- Apache::MethodList was implemented but was not usefull enough.
new modules:
- use apache '1.3b3';
misc new stuff:
- 'make html'
- 'make test_report'
apache features that would be neat for mod_perl:
- "autoload" hook for configuration directives
tools that may help us develop/maintain mp2
- B::Xref
- gcov: test code coverage
- valgrind (currently valgrind chokes when running apache, long before
we get to mod_perl). I've heard that we could use some special
#define controls to tell valgrind where to look and where not.
- gprof
- lxr
- tinderbox