[Bug pch/95131] New: Instantiate templates at pch generation time

[Bug pch/95131] New: Instantiate templates at pch generation time

[trass3r at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95131

            Bug ID: 95131
           Summary: Instantiate templates at pch generation time
           Product: gcc
           Version: 8.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: pch
          Assignee: unassigned at gcc dot gnu.org
          Reporter: trass3r at gmail dot com
  Target Milestone: ---

Judging by the limited information -ftime-report provides it looks like gcc
still spends quite some time on things that could have been done at pch
generation time already like template instantiations.

This was already noted in
https://gcc.gnu.org/wiki/Speedup_areas#Improvements_to_PCH
The page should be updated if this is not true anymore.

Clang has a similar problem that is currently being addressed:
https://llunak.blogspot.com/2019/11/clang-precompiled-headers-and-improving.html
https://reviews.llvm.org/D69585

[Bug pch/95131] Instantiate templates at pch generation time

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95131

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |nathan at gcc dot gnu.org

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Modules are the future, not sure how this applies there.

[Bug pch/95131] Instantiate templates at pch generation time

[nathan at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95131

Nathan Sidwell <nathan at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #2 from Nathan Sidwell <nathan at gcc dot gnu.org> ---
1) The future is sooner than you think :)

2) template instantiation cannot be done speculatively -- only when language
constructs require it.  Otherwise you can change the meaning of well-formed
code.

[Bug pch/95131] Instantiate templates at pch generation time

[trass3r at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95131

--- Comment #3 from Trass3r <trass3r at gmail dot com> ---
Well sure but most existing codebases won't be ported to modules because of the
effort while enabling pch is now straightforward with latest CMake.

These conformance issues have also been identified in
https://reviews.llvm.org/D69585#1946765,
which is why a flag -fpch-instantiate-templates was suggested.

For C++20 header unit compilation this behavior is actually required it seems,
so at least if -fmodule-header reuses pch code this may be a low-hanging fruit.


Is anything else from the wiki valid?
"Mangle names, instantiate templates, and generate debug information at pch
generation time"