[ts-gen] The shim now requires order keys ...

Bill Pippin pippin at owlriver.net
Thu Jul 8 12:22:29 EDT 2010


With respect to:

> ... thanks a lot for your ... emails ... cover all issues ...

You're welcome, and thanks much for the feedback.  Thanks also
for your reporting:
> I'll run ... the same 'order' script with the shim-100630 version
> and if the segfault still persist, I'll try to 'strace' ...  I'll
> keep you informed.

I'll be glad to get your report; I expect that it will enable me to
block off an unhandled error case somewhere.  Although I try to cover
all bases, it's actually quite difficult to check all possible error
cases without also having redundant checks.  The type system helps a
lot, and I use both the C++ language type system and an application
domain type system of first order and meta non-terminal language
symbols for just that reason, yet even so, experience shows me that
I still overlook some cases.

As long as you can get a reproducible result, the problem can be
fixed, one way or another.  I appreciate being able to leave it to
you, since I don't have a local reproducer.  Keep in mind, however,
that there is a known, already fixed parse loop for the 100630
tarball that might conceivably, under some circumstances, blow the
stack.  In that event, your report will just confirm that the new
release should supplant the old one.

More generally, you may well find a problem that only shows up with
older shims, as the tighter token checking for the newest release
prevents the dangerous code path from being taken for your test.

In that case, it will be well-worth it to me to fix such a problem
if the critical code point is not properly guarded, since experience
shows that I or someone else will probably stumble over it in the
future via some other approach if it's not fixed now.



More information about the ts-general mailing list