RE: [PATCH] Factorize the two read_line functions present in gcov.c and input.c

[Bernd Edlinger <bernd dot edlinger at hotmail dot de>]

> On 11/07/13 09:32, Dodji Seketeli wrote:
>
>> From the above, what I can say is that input.o was already linked with
>> gcov. But I think it's minimal enough to only drag libcpp and the
>> diagnostic subsystem.
>
> I disagree. While input.o was available to gcov, I don't think it was being
> pulled into the executable (if it was, I'd like to understand why). I think
> dragging in libcpp & diagnostics is a major bloat -- have you measured the
> before and after gcov text sizes?


I agree that this should be avoided. However both versions of read_line
need to be fixed eventually.

Just for the records, this is not only a theoretical issue for gcov:
The OOM error can also be reproduced in gcov.

The attached test.c has some dozends of comment lines:
//xx\0EOL

That are lines with text before and after the \0.
If compiled without -Wall, it will crash in gcov:


gcc -fprofile-arcs -ftest-coverage test.c
./a.out
gcov test
File 'test.c'
Lines executed:100.00% of 1
Creating 'test.c.gcov'

gcov: out of memory allocating 1677721600 bytes after a total of 135168 bytes


Bernd.

Attachment: test.c
Description: Text document