David> I did a fresh update of Cygwin and then downloaded the David> 21.1.14 tarball. Configure was run without any options. David> It failed to find XPM because the first of the two test David> programs would not compile because X11 header files David> could not be found. Configure for 21.2.43 does David> correctly detect XPM. David> I don't understand configure well enough to create a David> patch, but the only difference in the XPM section David> of the two configure scripts is that the 21.2.43 David> version has the line "#define XPM_NUMBERS" at line 7543 David> of the file. This would be at line 6958 of the 21.1.14 David> version. I would like someone in Windowsland to look at this more deeply. The fundamental principle of running test programs at configure time is to use external interfaces in the _same_ way as the real build process later does. Otherwise you have mismatches between what is detected and what actually works. I have worked to eliminate such mismatches over the years. I would like #define XPM_NUMBERS to be removed from the 21.2 configure script. Find another way to fix this in configure.in.

Executive summary: I reluctantly agree to add the #define XPM_NUMBERS to be added to 21.1. Craig> Martin Buchholz writes: Craig> The configure script does two checks related to the xpm library. The Craig> first one just wants to know what version the header file thinks is Craig> current, what version the library actually contains, and whether the Craig> two match. The second check is to determine whether FOR_MSW is needed Craig> or not. '#define XPM_NUMBERS' allows the configure script to check Craig> the existance and version of the xpm library without having to know Craig> whether FOR_MSW is needed. I looked at the detection in configure.in and I understand better now. Note that the current code does not check that xpm will actually work if FOR_MSW is defined, only that it fails if it's not. This is a (small) bug. However, fixing this is a little too risky for 21.1 or 21.4.0. I will add a comment to configure.in instead. I approve the suggested change to 21.1.15 configure.in. Someone who really fixes this should have access to both windows and unix. Craig> A windows build must define FOR_MSW before using the xpm library. Craig> When building with VC++ it is a non-issue because that build mechanism Craig> doesn't use 'configure'. However, the cygwin and mingw (-mno-cygwin) Craig> build mechanisms do and at least the mingw mechanism needs that define Craig> in order to properly detect the xpm library. Would it be unreasonable to simply AC_DEFINE(FOR_MSW) when window_system is "msw"? The tricky thing is using X Windows on MS Windows.