Andrew Black wrote:
Martin Sebor wrote:
Andrew Black wrote:
[...]
Attached is a patch that implements the rwtest and makefile changes
for this option. One problem observed is that the
22.locale.time.get, 22.locale.time.put, and 27.objects now report a
FORMAT status. The cause of these messages is output after the
assertion count has been printed. The compatibility mode parsing
logic searches for the summary total in a different way than the
normal mode, and is therefore affected by this output where the
normal mode isn't.
So is the problem that we start parsing too late in the file
(i.e., seek not far enough back) and end up skipping part of
the important output? Could we start parsing the output at
an earlier point in the file?
That would be an accurate assessment. It would be possible to move the
seek position to a spot earlier in the file, but it would also be
necessary to alter the FSM used to search for the assertion count, as
the current method searches for the string "\n## ", which will also
match in the header of the file. This spurious matching is what lead to
the change to output.cpp embedded into the patch.
Why do we seek in compatibility mode instead of processing
the file from top to bottom like we normally do? There could
be any number of notes or other garbage at the end of the
output file so seeking back is never going to be reliable.
Martin
