Update Windows build and add Windows to the CI lineup

[dag-erling]

This is a work in progress. Visual Studio reports a heap overflow in retest, so I'm not running tests in CI until I've found and fixed the cause.

[dag-erling]

I believe the heap corruption comes from retest setting use_regnexec set to 1 despite HAVE_REGNEXEC being undefined (because win32/config.h does not define TRE_VERSION). This results in a buffer overflow when wrap_regexec() appends a terminating null character to a buffer that does not have room for one.

[trushworth]

These all look good to me, with two minor questions.

1. I have an ancient (pre-ANSI-C) minimal implementation of getopt() that I use for windows stuff. It uses <stdio.h> (for fprintf(stderr,...) and <string.h> (for strchr()). If I clean it up to the GNU style conventions would it make sense to have it in the "tests" directory (or maybe the "win32" directory) for retest and test-str-source? We'd use the system getopt() when available obviously, but I'd like a working "-o" option on Windows. Adding it would probably be best as a completely separate pull request.

2. It would be nice to test the return code of tre_regcomp(). Testing that (in my own repo) is what eventually led me to finding the missing {}s.

I have some further additions that I would like to see, but I think the best way to do them is commit these first and let me pull them to my repository for merging and eventual inclusion in a future pull request.

[dag-erling]

I've added error checking for tre_regcomp() to test-str-source. I've also added an error message when make_str_source() fails.

[GerHobbelt]

Re your getopt for a windows platform:

On Fri, 13 Sep 2024, 01:54 trushworth, ***@***.***> wrote: ***@***.**** approved this pull request. These all look good to me, with two minor questions. 1. I have an ancient (pre-ANSI-C) minimal implementation of getopt() that I use for windows stuff. It uses <stdio.h> (for fprintf(stderr,...) and <string.h> (for strchr()). If I clean it up to the GNU style conventions would it make sense to have it in the "tests" directory (or maybe the "win32" directory) for retest and test-str-source? We'd use the system getopt() when available obviously, but I'd like a working "-o" option on Windows. Adding it would probably be best as a completely separate pull req May I suggest https://github.com/kimgr/getopt_port as a very nice getopt

+getopt_long replacement instead? It's BSD licensed so nicely unburdened that way too. I've used my own getopt implementation for years (decades, really) on Win32 and other platforms but switched to using this one a few years ago and it hasn't failed me since. git submodule add or git subtree add... :-) Thanks for maintaining tre and all the recent activity. Respectable.