bug#48497: 28.0.50; Native compile: excessive recompilation

[Andrea Corallo]

Michael Welsh Duggan <mwd@md5i.com> writes:

> The problem is that it seems as if standard Emacs lisp files from the
> installed base Emacs lisp library are getting recompiled in the user's
> emacs eln cache despite there being an existing native-compiled version
> installed.
>
> First, I did a fresh out-of-tree build from master using the following
> procedure:
>
> The git worktree is in a directory called "master".
>
> cd master
> ./autogen.sh
> mkdir ../build
> cd ../build
> mkdir -p ~/tmp/emacs/config
> ../master/configure --without-toolkit-scroll-bars --with-x-toolkit=lucid \
>   --with-native-compilation --prefix=~/tmp/emacs
> make -j 32 NATIVE_FULL_AOT=1 bootstrap
> make NATIVE_FULL_AOT=1 install
>
> I verified that there was a directoy,
> ~/tmp/emacs/lib/emacs/28.0.50/native-lisp/28.0.50-6152138a,  which
> contained 1382 .eln entries.
>
> I then started emacs like this, after first making sure there is no
> "~/.emacs.d" directory or symlink in my home directory:
>
> XDG_CONFIG_HOME=~/tmp/emacs/config ~/tmp/emacs/bin/emacs -Q
>
> At this point, ~/tmp/emacs/config was empty.  I then typed `C-h i',
> starting info.  This started some native compilation in the background.
> After waiting for the compilation (viewed in top) to settle, there was a
> directory, ~/tmp/emacs/config/emacs/eln-cache/28.0.50-6152138a, which
> contained the following .eln files:
>
> bytecomp-12882072-1f232d0e.eln  comp-cstr-ef162ef7-c492cd97.eln
> byte-opt-9c5f25f5-26e670d7.eln  gv-e0cf7478-ce8290a8.eln
> cconv-3b1f1f98-ac4628a4.eln     help-mode-d4dbae3d-034e8cf5.eln
> cl-extra-d7051cba-ff815ded.eln  info-ce12c0ca-3d2b00fd.eln
> cl-lib-8b938900-fbcee53d.eln    rx-627d8c83-18b6598f.eln
> cl-macs-7ae82f81-bc4ef696.eln   seq-c314b42f-4e726806.eln
> cl-seq-f464c4e2-95d0eb23.eln    subr-x-02dfef32-04f2924b.eln
> comp-7672a6ed-d617b299.eln      warnings-28e75f4d-c98d4e00.eln
>
> I verified that versions of these files existed in
> ~/tmp/emacs/lib/emacs/28.0.50/native-lisp/28.0.50-6152138a, which they
> did, but the first hash differs in each case, and the file sizes are
> completely different.
>
> Hopefully this is enough information to recreate the behavior.

Thanks it is, I can reproduce this.

  Andrea