[cfe-dev] -ferror-limit omits part of last error message

[Nikola Smiljanic]

On Thu, Aug 11, 2011 at 7:07 PM, Douglas Gregor <dgregor at apple.com> wrote:

>
> This doesn't look right… it looks like it'll allow *all* notes to be
> printed, even if the warning/error preceding that note was suppressed. What
> we really want is for the notes corresponding to the last displayed error to
> be displayed, but subsequent notes should still be suppressed.
>
> - Doug
>


You're right. It seems that this logic is already in place but it doesn't
work and here's why: The error is emitted and delayed diagnostic is set
(fatal error). So when ReportDelayed() is called it emits the fatal error
but it still has delayed diagnostic set (itself) which causes another
ReportDelayed to be called, and this one defeats the logic.

The fix would be to reset DelayedDiagID before reporting it, like this:

// store the object returned by Report in order to make its destructor run
after DelayedDiagID is set to zero
void Diagnostic::ReportDelayed() {
  DiagnosticBuilder diagBuilder = Report(DelayedDiagID);
  diagBuilder << DelayedDiagArg1 << DelayedDiagArg2;
  DelayedDiagID = 0;
}

// reset DelayedDiagID before calling Report, but create the temporary to
save the value
void Diagnostic::ReportDelayed() {
  int temp = DelayedDiagID;
  DelayedDiagID = 0;
  Report(temp) << DelayedDiagArg1 << DelayedDiagArg2;
}

// pass DelayedDiagID as the argument (this is passing its own member as an
argument)
void Diagnostic::ReportDelayed(int id) {
  DelayedDiagID = 0;
  Report(id) << DelayedDiagArg1 << DelayedDiagArg2;
}

The first one is the most reasonable one for me. What do you think?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20110812/909e2193/attachment.html>