On Tue, Oct 19, 2004 at 04:52:28AM -0400, Daniel Veillard wrote:
This rings a bell, but ... The closest I could find in the ChangeLog was a commit Tue Dec 30 16:26:13 HKT 2003 William Brack but this was before 2.6.6 . I think I also fixed some reuse leak when I made sure XInclude parsing shared the document dictionnaries, the fix may have been included as part of this. The key part is to be sure the reuse is tested as part of the regression tests the best way might be to isolate and drop a small example like the ones in doc/examples, this is both educational and make sure memory debug is exercised as part of the regression and release tests.
Ok I've had a go at an example, which is attached. It should either demonstrate the problem or my ignorance of the API :) The output could probably be made more useful for regression testing purposes. If the example is run linked against 2.6.6 some memory is leaked, if using 2.6.14 nothing is leaked. I've attached some Purify output for the 2.6.6 run. Interestingly, my original test case used xmlReaderForMemory and in that case Purify showed the memory as definitely leaked (MLK), but in this case it shows most of it as PLK (potentially leaked). cheers, Graham. -- Graham Bennett
Attachment:
reader4.c
Description: Text Data
Attachment:
purify.out
Description: Text document