GSoC Status: Week 6 and Half-Term Summary

[Matthew Windsor]

Hi there,

The last soc-status report of the half-term and before I go on a slightly
badly timed but pre-warned-in-proposal holiday is here.

Most of the changes brought in this week are sorely untested and the last
few might have broken off a few features that worked previously; as I said
in the last commit I did, I'll have a look on Monday to see if anything can
be quickfixed before the mid-term evaluation is up.

Null-pointer asserts have been added into most of the existing code
(r255055, r255056, r255077, r255099); obviously these aren't major
improvements but hopefully they've instilled me with a more proactive
outlook to adding in assertions in future and might trap some of the more
stupid bugs I'm introducing as I go along.

I wrote a very quick shell-script test for SearchFile (r255100) that
compares its output to that of pkgng.  More comprehensive testing is long
overdue and something I definitely want to touch on in the next term.

In r255152 I started working on UpdatePackages, but due to a
misinterpretation of the pkgng API what I actually implemented here was
UpdateSystem... which I then got confused with UpgradeSystem which is a
different PackageKit action involving distribution releases.  (Argh!)
r255153, r255127, r255231 and r255240 continue this confusion.
UpdateSystem hasn't been tested much (mainly because of it failing and the
events system not picking up the error to report it correctly), but in
previous pseudo-UpdatePackage issues it did seem at least to be solving
correctly.

At r255231 I finally started using full-length commit messages, as I
realised while looking at svnweb that mine were below average in terms
length and detail.

>From r255264 and including r255265, r255266 and 255285, I've been working
on replacing the query-based job resolution (as in, each action ending in a
job goes through query/rquery first and then pours the resulting packages
into their own jobs) with a system based on making one job and then
checking its solution for PackageID failures.  In the process, I've redone
the current job-based actions (InstallPackages/UpdateSystem/RemovePackages)
to use the new job system and also created UpdatePackages.

This is the source of quite a bit of breakage in the current system, I
fear, as it has been slightly rushed in order to have it done for this
status update.  I intend to give it a once-over tomorrow and do some
informal fix-ups.

Unfortunately I didn't get around to making the backend reject PackageIDs
with no version, but this will probably be done tomorrow.

So, at the end of the half-term, the backend has a lot of code implemented
covering the core features of
install/update/remove/search-by-name/get-details-of-package, and many of
the common patterns (query-based actions, job-based actions, package
manipulation, packageID manipulation) have been identified and abstracted.
What remains really for the next half-term is to squish bugs (and there are
*many* bugs, some of them show-stopping), round out the remaining
functions, polish the code and ready it for a ports release with pushing to
PackageKit as a possibility once it's ready.

Finally, here is what I think will be the rough, minimal outline of work
for next term (I'll have a more detailed overview before I head off on
Tuesday):

(Week: Goal)
6: Quickfixes on week 5 implementation
7: Remove query-based jobs; improve events system to provide correct errors
on job failure; more in-depth checking for any issues remaining from the
jobs change.
8: GetDepends, SearchDetails and GetRequires; at this stage with the
exception of Cancel the backend should be at feature parity with ports.
In-depth attempts to fix the install crash bug.
9: Try implementing Cancel, RepoEnable (text file hackery probably), and
WhatProvides.  Otherwise, tests.
10: Tests for each idempotent action.
11: Tests, where possible, for non-idempotent actions, jail/VM if possible.
12: Create a port and upload to ports tree.  Also, final call for updating
PackageKit: if possible, updating; if not, contingency and cleanup.
Between soft and hard deadline: Contingency

~ Matt