Dear Paul, On 07/21/2011 01:25 PM, Paul Richard Thomas wrote:
We are making increasing use of scan-tree-dump-times in the tests. I wonder if this is well advised?
I think it is the only way to make sure that things work as advised - the other possibility is to look at the assembler output: scan-assembler and scan-assembler-not.
Doing things just on the basis whether it compiles is not as reliable - and run tests are much more expensive. Regarding the latter: I think it would be useful to have a run-once test mode. This mode is useful for testing the library and some correctness issues without being as heavy as a full blown run with different optimization options.
I have been hit in the past by subsequent patches affecting these tests even though the relevant patch is working fine. That said, it will flag up the changes if and when array descriptor reform is finally done.
I concur that there are issues such as platform dependence (I wouldn't be surprised about such failures for this patch) and it makes it also more difficult if one does more invasive patches, which affect many test cases without changing their result.
But I do not see any good alternative. Looking at dumps while writing patches has often found problems, which exist but do not show up at run time. For instance, the initial patch (w/o the se->wants_pointer part of the patch) didn't simplify the offset to 0 if one passed the same object. And similar issues can be introduced when working on other patches - and only such scan tests can find them.
Thanks for reviewing the patch, which has been committed as Rev. 176562. Tobias
