Re: CHECK_STRUCTS/dmpstruct.h mechanism is broken.

[Andy Moreton]

On Mon 15 Apr 2019, Andy Moreton wrote:

> On Wed 10 Apr 2019, Paul Eggert wrote:
>
>> Daniel Colascione wrote:
>>> What would make the hashes easier to deal with? Some make target for
>>> updating them automatically? An easier way to disable the check at
>>> configure time? Transformation into a prominent runtime warning instead of
>>> a build break?
>>
>> It's better to have the build break than to have a runtime check. Of the two
>> other approaches you mentioned, I think a configure-time option is the better
>> idea. Since these hashes are aimed at developers and are are not needed for
>> ordinary builds from tarballs, I think the hashes should be checked on
>> request. I resurrected the hashes by installing the attached patch and they
>> are now checked with the longstanding 'configure' option 
>> --enable-checking=all
>> or with the new 'configure' option --enable-checking=structs. Comments 
>> welcome
>> as usual.
>
> This newly resurrected checking is already broken on master, which does
> not build with confgiure option "--enable-checking=yes,glyphs".
>
> The struct fingerprint checking should default to enabled (as before)
> with the configure arguments allowing it to be (temporarily) disabled.

Apologies - the problem is actually in commit 5c2f94a182 ("Replace
executable’s fingerprint in place"):

  CCLD     temacs.exe
C:\emacs\git\emacs\master\build\mingw64-x86_64\lib-src\make-fingerprint.exe: 
Error: temacs.exe.tmp is not a regular file
make[1]: *** [Makefile:644: temacs.exe] Error 1
make[1]: Leaving directory '/c/emacs/git/emacs/master/build/mingw64-x86_64/src'
make: *** [Makefile:423: src] Error 2

I expect this is something involving the libc handling for Windows
builds, as the fstat() call reports st.mode as 0 rather than continaing
the expected 0x8000 bit for a regular file.

    AndyM