[cfe-dev] Downgrading/mapping fatal errors to errors (#include not found)

[Richard Smith]

On Sun, Jan 4, 2015 at 8:15 PM, Jason Haslam <jason.haslam at gmail.com> wrote:

> I wonder if it should continue to report all fatal errors instead of just
> file-not-found.
>

I think it makes sense to report file-not-found repeatedly, because
recovery from previous errors is very unlikely to be the cause of later
file-not-found errors. For the other fatal cases, that is not the case, so
I don't think it makes as much sense to suppress them.

Also, my patch sort of circumvents the 20 error fatal cutoff. Maybe that’s
> okay.
>

That seems problematic; if essentially-correct code #includes a bunch of
headers, but the build fails due to a missing -I path, it seems especially
important to respect the error limit.

I would prefer that we add a flag to diagnostics to indicate if they should
be shown even after a fatal error, and then mark err_pp_file_not_found with
that flag in the .td file, rather than hard-coding that particular
diagnostic.


> Jason
>
>
> On Jan 4, 2015, at 2:19 AM, Richard Smith <richard at metafoo.co.uk> wrote:
>
> On Sat, Jan 3, 2015 at 11:15 PM, Jason Haslam <jason.haslam at gmail.com>
> wrote:
>
>> This is important to us for the same reasons. In our case we only care
>> about reporting additional file-not-found errors.
>
>
> Olivier, is this also the case for you, or do you want to see other
> diagnostics after a missing #include?
>
>
>> The others usually become too noisy. I haven’t found any existing option
>> to control this. We have a local patch to continue reporting file-not-found
>> errors after a fatal error (may be slightly out of date):
>>
>> --- a/lib/Basic/DiagnosticIDs.cpp
>> +++ b/lib/Basic/DiagnosticIDs.cpp
>> @@ -618,7 +618,7 @@ bool DiagnosticIDs::ProcessDiag(DiagnosticsEngine
>> &Diag) const {
>>
>>    // If a fatal error has already been emitted, silence all subsequent
>>    // diagnostics.
>> -  if (Diag.FatalErrorOccurred) {
>> +  if (Diag.FatalErrorOccurred && DiagID != diag::err_pp_file_not_found) {
>>      if (DiagLevel >= DiagnosticIDs::Error &&
>>          Diag.Client->IncludeInDiagnosticCounts()) {
>>        ++Diag.NumErrors;
>>
>> I would love to see an option to control this behavior added to the
>> mainline.
>
>
> I don't think we need an option for this; it seems reasonable to always
> report missing #includes, even if we've already hit a fatal error, because
> a missing include is very unlikely to be caused by a prior missing include
> (they may have a common cause, but that's OK).
>
> Are there already other flags meant specifically for IDE integration?
>> Maybe this could piggyback on one of them or do something similar?
>>
>> Jason
>>
>>
>> > On Dec 31, 2014, at 4:13 AM, Olivier J. G. <olivier.jg at gmail.com>
>> wrote:
>> >
>> > Hello,
>> >
>> > We've run into a problem using libclang for cpp support in KDevelop:
>> once an #included file is not found, a fatal error is generated which
>> disables all further diagnostics.
>> >
>> > In the context of KDevelop, it is quite important that this be
>> downgraded to a non-fatal error so that further missing includes and issues
>> can be marked in the IDE. It's further desirable that the IDE support is
>> still helpful in the absence of a missing but unimportant #include.
>> >
>> > I've found no way to do this with compiler flags; err_pp_file_not_found
>> has no category so -Wno-fatal-errors=foo can't help (not clear if that's
>> the correct method even if there was a category).
>> >
>> > Is there a way to do this that I missed seeing? Can this diagnostic be
>> re-mapped from fatal to error?
>> >
>> > If there's no existing way to do this, could some interface or compiler
>> flag be added so that err_pp_file_not_found_not_fatal is used instead? What
>> would be the correct place/way to configure this?
>> >
>> > Thanks,
>> >
>> > -Olivier JG
>> > _______________________________________________
>> > cfe-dev mailing list
>> > cfe-dev at cs.uiuc.edu
>> > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>>
>>
>> _______________________________________________
>> cfe-dev mailing list
>> cfe-dev at cs.uiuc.edu
>> http://lists.cs.uiuc.edu/mailman/listinfo/cfe-dev
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20150105/2b52a712/attachment.html>