On 3/25/19 5:28 AM, Paolo Carlini wrote:
Hi,
a while ago I noticed that in cp_parser_binary_expression we were
calling cp_fully_fold first to disable the warnings for not executed
expressions, and then, seemingly in a very redundant and inconsistent
with other situations way, to re-enable the warnings (*). Thus, I meant
to investigate opportunities for a mini optimization / clean-up during
the next Stage 1.
However, over the last couple of weeks I noticed that we had in Bugzilla
a number of error-recovery regressions happening starting from the
*second* cp_fully_fold calls, those which likely could be completely
avoided. Note, not only we were calling again cp_fully_fold, we were
also doing that for expressions which could not be folded to true/false
the first time we tried, error-prone and wasteful too.
Then the below which seems rather straightforward to me. To be
super-safe I carried out a number of additional checks and instrumented
testsuite runs: that we exercise the code enough; that when we set
disable_warnings_sp = sp we find it NULL, etc. Everything went well..
Tested x86_64-linux.
Thanks, Paolo.
(*) Historically, we used not to have this disabling code at all, then a
very basic version not folding.
/////////////////
OK.
Jason
