Re: [patch] initial location support for C parser

["Joseph S. Myers" <joseph at codesourcery dot com>]

On Fri, 22 Aug 2008, Aldy Hernandez wrote:

> Joseph (and Jason), I would like your input on this before I go on.  I
> am committed to fixing the location information throughout the compiler.
> And though I know you like to have everything fixed in one go, I work
> better with an incremental approach.  I sincerely hope you can
> accomodate this.
> 
> I am going through tests in dg.exp and verifying that we have the
> correct column information.  I'd like to fix as I go.
> 
> With this initial patch I noticed that to come up with suitable tests
> for all this, I would have to basically write an entire ISO C
> comformance test, something which is perhaps beyond the scope of my
> work.  I would like to fix existing tests as I go, and perhaps add new
> ones when it is obvious how to add new tests.  Sometimes I noticed it
> was incredibly hard to contrive a test for a particular location we were
> setting.  I hope we can come to a middle ground here-- I doubt I have
> the time, desire, or inclination to write an entire comformance suite.

The C parts of the patch are OK.

In general, I think the patch changing a diagnostic that relies on the 
implicit input_location to take a more specific location is probably the 
patch that should have an assertion in the testsuite about exactly what 
(including column) that location is; for patches such as this that set 
locations it's not at all clear what the relevant diagnostics, if any, 
are.

When I was expanding test coverage for C front-end diagnostics to avoid 
the new C parser breaking untested areas of the compiler, I used a script 
to identify messages present in a newly generated gcc.pot but not in 
gcc.log from a testsuite run; this is attached.  It's obviously not quite 
what you want for this work, but should illustrate how, given a message 
from the compiler, to find testcases that exercise that message so you 
know what tests to modify to add column number tests for a particular 
message.

-- 
Joseph S. Myers
joseph@codesourcery.com

Attachment: pot2msgs
Description: Text document