195608 - tinderbox warnings parser doesn't handle multiline warnings

Status: RESOLVED WONTFIX

(Webtools Graveyard :: Tinderbox, defect)

Description

[Bradley Baetz (:bbaetz)]

Comment 1

[Bradley Baetz (:bbaetz)]

Err, with a comment this time.

gcc >=3 prints warnings on multiple lines. The tbox warnings parser can't handle
that, and so only prints the first part of the line.

I've made a patch to fix this, but its a bit hacky - see comments in the patch.
Except for the possible mingw issue, which may not be an issue if hat uses
cygwin paths, its no more hacky than the rest of that file.

This produces the same results on myotonic (not running gcc-3), and the same
number of warnings on brad if I disable the stripping of 'hidden' warnings and
some of the @ignore stuff. (I had to do that because they didn't match in the
original code, because the string was stripped).

I had to munge some stuff to get this testing since I don't actually have  atbox
server (eg hard coding gcc_parser), but thats not in the patch.

Comment 2

[Bradley Baetz (:bbaetz)]

Attachment: fix

I keep $line arround as a possibly-read previous line. This is needed because I
have to pull out the next line before I know if its a multiline warning, and
just ignoring it misses the second warning in an immediately adjacent pair.

Comment 3

[tenthumbs]

Wouldn't it be better to use the -fmessage-length=0 compiler option 
where appropriate rather than depending on gcc's message structure which 
could change at any time?

Comment 4

[Bradley Baetz (:bbaetz)]

Yes, but that requires the tinderbox owner to set that option. (It can't be
forced in the tinderbox scripts because older gcc's won't understand that option
and will presumably error out) I don't know if thats considered an issue, though.

I also read the text 'Right now, only the C++ front end can honor these
options.' as meaning that only c++ will honour a request not to wrap, not that
only c++ does any wrapping. Quick testing shows that I was wrong, so this would
work.

If someone wants to WONTFIX this, then I'll change the CFLAGS/CXXFLAGS on the
Brad tinderbox

Comment 5

[tenthumbs]

Yes, gcc 2.95.x gives up on options it doesn't know about so you have to 
be a little careful. I've been using -fmessage-length=0 on my own builds 
for quite a while and it works so the gcc and g++ front ends can handle 
it.

Comment 6

[Christian :Biesinger (don't email me, ping me on IRC)]

bbaetz: My gcc info page says:
`-fmessage-length=N'
     Try to format error messages so that they fit on lines of about N
     characters.  The default is 72 characters for `g++' and 0 for the
     rest of the front ends supported by GCC.  If N is zero, then no
     line-wrapping will be done; each error message will appear on a
     single line.

In other words: By default, only g++ does wrapping, gcc doesn't.

Comment 7

[Bradley Baetz (:bbaetz)]

Yeah, as I said, I misread the docs the first time.

Comment 8

[Chris McAfee]

If someone wants to try the patch out on tinderbox, I can help do that today,
lemmeknow aim:sleestack8.

Comment 9

[Bradley Baetz (:bbaetz)]

timezones mean that thats not likely to happen, but if you install it, then the
warnings for brad should change, but nothing else.

That assumes that this patch is wanted, mind you...

Comment 10

[Sylvestre Ledru [:Sylvestre]]

Tinderbox isn't maintained anymore. Closing.