I think the first point of confusion stemmed from those bash CGI scripts you mentioned. I could have sworn I saw more shebangs with paths in /usr/bin/, but that might have come from my semi-odd setup. I'm ssh-ing into cygwin then running the test suite with Strawberry perl through a couple of batch scripts I setup; I probably mistakenly ran Makefile.pl using cygwin perl at some point. Just to clarify, the server is built using the visual studio compiler, not using MinGW or Cygwin gcc, which is why I'm not using cygwin perl.
Thanks for the link to the modperl apxs; I think I saw references to it around the web, but all the links were dead. On Wed, Jan 21, 2015 at 1:02 PM, <wr...@rowe-clan.net> wrote: > --------- Original Message --------- > Subject: Running the test suite on Windows > From: "Edward Lu" <chaos...@gmail.com> > Date: 1/21/15 10:05 am > To: dev@httpd.apache.org > > It appears that the test framework has a bunch of dependencies on unix, > e.g. it runs scripts with "#!/usr/bin/...". Running it under cygwin would > fix some of those issues, but unfortunately, running under cygwin perl > generates unix-like paths in the conf files which a windows-built httpd > can't deal with. I'm currently trying to find where it configures those > paths so I can hack around the issue. > > There's also the problem of Windows not having apxs, so a bunch of tests > are skipped, but I'm ignoring that one for now. > > Does anyone have experience/advice/considerations to give on the subject? > > > I'm reasonably confused by your first statement and I'd need a bit of > clarification > to comment further. You can't mean forward-slashed paths, those are just > fine, > even by windows itself; just like passing -- before passing file names > with leading > dashes on unix, you do need to disambiguate the option flag for windows own > commands by enclosing such paths in quotes. From a windows command shell > prompt, you can observe this with the command dir "c:/windows" > > Not sure which other scripts you are describing; if specific tests use > script to > test a feature, e.g. a test to execute a bash shell cgi, then that would > simply be > "not applicable" to windows (usually). You could introduce further tests > that do > exercise all of the mechanisms in the win32 cgi functionality (use > registry verb > associations or a shebang). But if you are describing something else, it > must > be a more recent test; the entire suite was developed in perl. > > Compiling under cygwin (or a truly cygwin build with a faux-posix clib) > may have > some interesting implications. APR implements constructs in a very NT > kernel > centric way with object handles, so a posix-like cygwin build might become > confused at one point or another. I wouldn't expect a mingw build to have > such > issues since that is a wrapper over the msvcr runtime. If you can track > these > down, it would be nice to work around the problem cases. > > Finally, there is an apxs that was never integrated into the httpd > project, by our > friend Randy Kobes, and Jeff along with the modperl folk finally moved it > into > subversion (sadly, not httpd's yet :) > > https://svn.apache.org/repos/asf/perl/apxs/trunk/ > > Configure and build it on top of an already installed win32 httpd. >
