[Tcllib-devel] Status of release work (RE: (proposed) Schedule for Tcllib 1.9 (Re: anythoughts on tcllib release?))

[Andreas K. <and...@ac...>]

Having done testing during last weekend and weekend I believe that we are in a
reasonable shape for a release. This is however so far based only on testing on
a linux/intel 32 bit machine. Anybody, please grab the CVS head and run its
testsuite on the machines/OSs your care for, with the tclsh/wish you care about,
with, or without C accelerators (critcl).

Regarding that single machine I was using I believe that I managed to get a
complete survey, using wish 8.[2345], testing without and with the C
accelerators, and running the latter under valgrind as well.

After fixing a number of things the only testsuite failures I see are for 8.5a5
(CVS head yesterday afternoon PDT). The main stuff is changed precision in
floating point arith which the testsuites are not accounting for. The only real
bugs I see are for bigfloat, which apparently computes completely different
results under 8.5a5 than for 8.4. This is something Sarnold75 has to look into.

At the C level valgrind shows a number of problems, these are however
concentrated on Tcl/Tk 8.5 and something in the core, finalization issues we can
do nothing about here. The snit testsuite manages to induce the problem so
heavily that seg.faults happen, however valgrind shows that the invalid
read/write accesses in question happen even for a module like 'bench', which has
not testsuite, and definitely no other C packages involved.

A second set of valgrind warnings happens for package sha, under Tcl 8.4, and
the Trf accelerator package active. The problems seem to originate in Trf, sha
merely manages to induces it, albeit there is no crash. Again nothing Tcllib can
do about. I will have to investigate this, however I am defering that
investigation until after the release of Tcllib.


The next goal for me is to go over the 80 modules (and the 244 packages
contained therein), check the version numbers, straighten them out, write up the
README documenting the changes ... for a RC by Friday this means 16 modules/day,
and 49 packages/day avg.

I consider the probability of me running the testsuite on some other
machines/architectures during this week to be basically zilch.


I did some changes to the test command which I will commit tonight so that
others use the same framework I did. Not all the changes I wanted will be
committed. The plan was to have the test rns store the results in a
mini-database, however I haven't written the code for reading the results back
and formatting them into nice tables, and the db is also slow to read. This part
is therefore disabled. Even so the test system is usable, it just doesn't
accumulate results.


>   *	Release Candidate ready at  Friday Sep 22.
>
> 	2 weeks after the freeze.
>
>   *	General Announcement    at  Friday Sep 29.
>
> 	Another week of bughunting and cleanup work.
>
>   *	Conference		Mo-Fri     Oct 9-13.
>
> One thing I wanted to do, the small 'name server' based on comm and
> the changes
> to comm in the last few weeks, I have decided to defer. It is not fully stable
> yet IMHO and trying to get it done would be detrimental to the other release
> work I have to do.
>
> I will from now on concentrate on running the testsuites, fixing up package
> versions, etc.

--
	Andreas Kupries <and...@Ac...>
	Developer @ http://www.ActiveState.com
	Tel: +1 778-786-1122