This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [ALL TARGETS] final patch for PR 25241

On 10/07/07, Mike Stump <> wrote:
On Jul 10, 2007, at 9:32 AM, Manuel López-Ibáñez wrote:

-/* { dg-error "error: unterminated comment" }
+/* { dg-error "unterminated comment" }
-signed long _Decimal64 *x107; /* { dg-error "error" "signed long
_Decimal64" } \*/
-signed long _Decimal128 *x107; /* { dg-error "error" "signed long
_Decimal128" } \*/
+signed long _Decimal64 *x107; /* { dg-error "" "signed long
_Decimal64" } \*/
+signed long _Decimal128 *x107; /* { dg-error "" "signed long
_Decimal128" } \*/

I found them when ensuring that all edited lines were actual dg-
commands and not just some random text that accidentally hit.  The
first two are, well, weird, but ok, the last 4 I cringe at, but, these
aren't detractions from your patch.

Those things were there already. But I agree, it would be better to fix the \*/. Moreover, it would be perfect if the "" would be replaced by a significant part of the error message. Could you prepare a patch for that after my patch?

I think it would be reasonably safe to just include all edits to
testsuite, even if their drivers don't yet make the distinction.

I disagree. I did run the script for the whole testsuite directory (notice how it handles "Error" and "Warning" for gfortran), however, I reverted the changes for g++.dg g++.old-deja and gfortran.dg, because those don't make the distinction yet. The problem with applying the changes if the driver doesn't make the distinction is that a regression may go unnoticed (particularly on a test where the author wanted to make the distinction explicit).

However, this only applies to "automatic fixes", there may be
testcases broken by the patch that may require manual tweaking. In
particular, g++ seems to have a huge amount of that. In addition, the
diagnostic messages in gfortran are quite irregular and difficult to
parse. Therefore, it may take a while (from a few months to a few
years depending how many people volunteer to work on this) for their
drivers to make the distinction.

Thanks for looking at it,


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]