[PATCH] PR14795 : -frewrite-includes sometimes results in incorrect line number

[Eli Friedman]

On Sun, Jul 28, 2013 at 2:22 AM, Lubos Lunak <l.lunak at suse.cz> wrote:
> On Thursday 25 of July 2013, Eli Friedman wrote:
>> On Tue, Jul 23, 2013 at 12:15 AM, Lubos Lunak <l.lunak at suse.cz> wrote:
>> > On Tuesday 16 of July 2013, Eli Friedman wrote:
>> >> On Sun, Jul 14, 2013 at 2:16 PM, Lubos Lunak <l.lunak at suse.cz> wrote:
>> >> > On Monday 01 of July 2013, Eli Friedman wrote:
>> >> >> --- a/lib/Lex/Lexer.cpp
>> >> >> +++ b/lib/Lex/Lexer.cpp
>> >> >> @@ -2372,8 +2372,9 @@ bool Lexer::LexEndOfFile(Token &Result, const
>> >> >> char *CurPtr) {
>> >> >>      FormTokenWithChars(Result, CurPtr, tok::eod);
>> >> >>
>> >> >>      // Restore comment saving mode, in case it was disabled for
>> >> >> directive. -    resetExtendedTokenMode();
>> >> >> -    return true;  // Have a token.
>> >> >> +    if (PP)
>> >> >> +      resetExtendedTokenMode();
>> >> >> +    return true; // Have a token.
>> >> >>    }
>> >> >>
>> >> >>    // If we are in raw mode, return this event as an EOF token.  Let
>> >> >> the caller
>> >> >>
>> >> >> How is this related?
>> >> >
>> >> >  I had a crash there without this when I wrote the patch. I don't
>> >> > remember the circumstances though and the random testcase I tried
>> >> > doesn't trigger it.
>> >>
>> >> Okay.
>> >>
>> >> Please commit without this change.
>> >
>> >  Actually, LibreOffice is always a good testcase, so I can reproduce the
>> > crash after all:
> ...
>> >  So I will include that change too, unless somebody sees a better way of
>> > avoiding this.
>>
>> Can you reduce a testcase?
>
> $ echo -ne '#if 0\n#endif' >a.cpp
> $ clang++ -frewrite-includes -E a.cpp
>
> (note the lack of the trailing \n)

Thanks.

Isn't the right fix here to rearrange the checks in Lexer::LexEndOfFile?

Please submit this as a separate patch, with a test.

-Eli