#9 Request for debug-friendly 25.0.90 pretest version

[Kaushal Modi]

Hi Chris,

Thank you for providing the 64-bit Windows builds for the latest Emacs dev versions.

Someone had requested a Windows build for the emacs pretest 25.0.90 on the emacs-devel GNU mailing list and I pointed him to that pretest build by you. As per the feedback on that mailing list, this build crashed ( http://thread.gmane.org/gmane.emacs.devel/199039/focus=199091 ) and he was unable to provide a useful backtrace to help debug that.

Eli (a very active Emacs contributor) suggested on the debbugs thread ( http://debbugs.gnu.org/cgi/bugreport.cgi?bug=22502 ) that certain flags should not be used. Would it be possible for you to provide a new build with those suggestions?

Configured using:
configure --prefix=/tmp/emacs --without-dbus --without-imagemagick
'CFLAGS=-O2 -fomit-frame-pointer -g0' 'LDFLAGS=-static-libgcc
-static-libstdc++ -static -s -Wl,-s''
^^^^^^^^^
Whoever built that binary, stripped the symbols (see the "-s" switch
above), so any such reports are useless, as no one can ever do
anything with them. Too bad.

Would the person who is building these binaries please refrain from
using switches that are hostile to debugging, such as
-fomit-frame-pointer and -s?

If you have a strong reason to use those switches, can you please bring it up on emacs-devel@gnu.org so that people knowledgeable in this area can contribute to the discussion?

Lastly, would you consider contributing your 64-bit Windows builds to the official GNU Emacs builds location? That will greatly help the Emacs community who rely mainly on Windows.

Thanks.

Kaushal

[zklhp]

Hi Kaushal,

Thank you for raising this issue. I’m really sorry for the inconvenience. I’ve read the feedback made by you and Eli. As Eli said, these switches are obtained from the MSYS2 people. I’m now building the pretest with:

CFLAGS='-O0 -g3' ./configure --prefix=/tmp/emacs --without-imagemagick --enable-checking='yes,glyphs' --enable-check-lisp-object-type

And the new pretest has been uploaded (http://sourceforge.net/projects/emacsbinw64/files/pretest/emacs-bin-w64-25.0.90.7z/download). Could you tell me is it Ok?

Also, I don’t know whether the emacs-w64 can be the offical build, as I can’t guarantee the build can “work” without including the dlls.

Again thank you and Eli.

Cheers,
Chris

[Eli Zaretskii]

Thanks for using these switches. It is better to use "-Og -gdwarf-4 -g3" instead of "-O0 -g3", as that allows pretesters to test optimized code, while still keeping useful debug info in the binary.

[zklhp]

Hi Eli,
Thank you for your suggestion. A new version built with these switches is uploaded.

Thanks,
Chris

[Eli Zaretskii]

Thanks, much appreciated.

[Kaushal Modi]

Thanks for the update! I did not see your build crash for me. So I will share this in the debbugs thread where this crash was reported. Hopefully they should be able to recreate the crash and debug and uncover a legit bug.

Would you like me to add you to the discussion thread on emacs-devel to help understand if your builds can be used as official builds? I am not entitled to tell what are the requirements for anyone to contribute builds as official builds.

If you'd like that, what email can I use to copy you on?

[zklhp]

Hi Kaushal,

Please mail me at chriszheng99@gmail.com.

Regards,
Chris

[İsmail Dönmez]

Would it be possible to keep the old flags and provide a -debug build instead with -Og and such? I am afraid enabling -Og will decrease overall performance (gcc docs say -Og results in "reasonable performance" which is not encouraging).

[zklhp]

Hi İsmail,

I made some change to the name scheme. Please go to https://sourceforge.net/projects/emacsbinw64/files/pretest/ and see whether it is appropriate.

Thanks,
Chris

[İsmail Dönmez]

Works for me! :) Sadly I think it'll triple your work, so I appreciate all your work.

[Eli Zaretskii]

In my experience, your fears about performance are unbased. The -Og version works fast enough. I urge you to use that version because doing that will help us with testing the upcoming 25.1 release, which, after all, is the main purpose of having the pretest binaries available.
Once v25.1 is out, the official release can be compiled with more aggressive optimizations, of course.

Thanks.