[Bug fortran/99053] New: memory leak in module reading, gfc_new_symbol

[Bug fortran/99053] New: memory leak in module reading, gfc_new_symbol

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99053

            Bug ID: 99053
           Summary: memory leak in module reading, gfc_new_symbol
           Product: gcc
           Version: 11.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rguenth at gcc dot gnu.org
  Target Milestone: ---

It looks like gfc_symbols as allocated via module reading are never
deallocated, even when the frontend finished parsing.  Compiling 521.wrf_r
module_cpl.fppized.f90 like

1612940030.01: ==2296== Memcheck, a memory error detector
1612940030.01: ==2296== Copyright (C) 2002-2017, and GNU GPL'd, by Julian
Seward et al.
1612940030.01: ==2296== Using Valgrind-3.15.0 and LibVEX; rerun with -h for
copyright info
1612940030.01: ==2296== Command:
/home/rguenther/install/gcc-test/usr/local/bin/../lib/gcc/x86_64-pc-linux-gnu/11.0.0/f951
module_cpl.fppized.f90 -I . -I ./netcdf/include -I ./inc -quiet -dumpbase
module_cpl.fppized.f90 -dumpbase-ext .f90 -march=znver2 -g0 -Ofast -std=legacy
-fconvert=big-endian -fno-openmp -fintrinsic-modules-path
/home/rguenther/install/gcc-test/usr/local/bin/../lib64/gcc/x86_64-pc-linux-gnu/11.0.0/finclude
-o /tmp/ccSQlHm7.s

shows this and various other fortran FE leaks.

[Bug fortran/99053] memory leak in module reading, gfc_new_symbol

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99053

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
If there's currently no part in the FE that tries to release memory of this
kind then a valid strathegy would be to change the current heap allocations via
X[C]NEW to allocations from a frontend specific obstack and simply release
the whole obstack at the point the FE transfers control to the middle-end
(assuming later langhook calls from it will not reach into the data, of
course).

Guess FE people will have an idea of the lifetime of objects.