From: Julien L. <ju...@fa...> - 2007-05-02 14:29:00
|
Keith MARSHALL wrote: > Julien Lecomte wrote, quoting me: >>> I've imported Tor's original package into CVS, and made a number >>> of changes to the build/packaging system. >> I've checked-out the package and ran a couple of tests, and have >> these following comments: >> >> - Under MSYS I get from 'make' this final output that looks like >> an error message to me, even if it's not. It's not very clear: >> "MSVC program 'lib' not available; not building MSVC components >> Not executing command: lib -def:libgnurx.def -out:gnurx.lib" > > That's my workaround, to avoid the even more catastrophic looking > `make: Error...' diagnostics thrown out by Tor's original, if you > don't have MSVC, (or at least a compatible `lib'), installed. It > isn't an error; just a note to the effect that no import lib for > MSVC is being built. If you think it isn't sufficiently clear, > can you suggest alternative wording? Perhaps > > --------------- > > Not executing command: lib -def:libgnurx.def -out:gnurx.lib > > MSVC program 'lib' not available; continuing without building > an import library for MSVC. This is not a problem, unless you > specifically require such an import library; if this is the > case, then you must install MSVC, and ensure that its `lib' > command can be found in your PATH. > > --------------- > I just also thought of the fact that I have "MAKEFLAGS=--no-print-directory" by default on my systems; thus maybe that's why it was less clear to me. Anyway, it's a MSVC specific warning message that's shown to MinGW users that don't use MSVC in that specific configure case. For me it would be clearer not to even have a warning. Did I mention I didn't really like MSVC anyway? ;-) >> - What's the real use of testing for 'tune=pentium3' in the >> configure file? or of mthreads for that matter? > > Tor's original Makefile specified these unconditionally. I started > out testing my modified build infrastructure, compiling natively on > GNU/Linux, where GCC chokes on `-mthreads'. Since I already had the > autoconf macro to test for portability of CC options, I used it; it > lets the native compile on GNU/Linux go one step further. > > Of course, it then falls over on the the next step -- DLL creation, > so it didn't buy me much. I guess it would be cleaner to rip it out > again, and leave these as unconditional addenda to the CC definition. I don't see where regex needs threads anywhere anyway. >> Under MinGW, compilation fails with missing ctype reference, so I >> guess regex isn't suppose to be compiled for MinGW. > Are you saying that the compilation of gnurx-0.dll itself fails? ... > Or are you saying that you can't deploy it with your own programs? I found the reason that compilation failed under MinGW for me: - I had first issued a './configure --prefix=/usr && make" under a MSYS shell in my regex src dir (/var/src/regex) - I then issued a './configure --prefix=/mingw && make' under MinGW shell Since I didn't issue a 'make clean' in the MinGW shell, "make" thought that the objects were up to date even if config.* files had changed. I believed that if config.h had changed, then "make" would be smart enough to recompile objects; at least that's how it works with 'automake' ;> Julien |