[Bug target/64368] [5 Regression] Several libstdc++ test failures on darwin and others after r218964.

[hp at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64368

--- Comment #11 from Hans-Peter Nilsson <hp at gcc dot gnu.org> ---
Created attachment 34344
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34344&action=edit
Call-trace for testsuite/22_locale/locale/cons/6.cc on cris-elf, a "newlib
target"

Plain execution trace from 22_locale/locale/cons/6.cc compiled with default
testsuite options, so inlining and whatever effects apply.

In the trace, lines with ">" denotes function calls, "<" are returns, with
indentation supposed to show the call-level.  The first four columns are the
hexadecimal values of the first four parameters as applicable (actually,
"parameter registers" but none are larger than 32 bits here, so same thing) /
ditto return values, padded with random contents.  (The fifth column is
cumulative execution times at each call-level.)

By a quick glance it seems the freed-but-not-allocated pointer, 0x5e23a, is
also the argument to the last call to strlen.

HTH.