GSoC status - Week 3

[Matt Windsor]

Hi all,

Another week, another soc-status mail!

This week I've mainly been implementing basic forms of the InstallPackages 
(remote installation) and InstallFiles (local installation) features.  The 
good news is that I've managed to install (trivial) packages with both, so 
they are mostly functional.

There are a few issues to iron out, most notably perhaps being the fact 
that I implemented events on InstallPackages a bit too overzealously and 
it now bombs out with a fatal error if mild things such as cached packages 
having checksum mismatches occur.  More cosmetic problems include the 
install queues being unsorted (and thus causing a large amount of "these 
packages are to be installed, ...reinstalled, ...installed, ...updated, 
...installed" noise).

I also implemented a very basic test, mostly for me at this stage, to 
check to see if GetDetails gets broken by any changes (and there have been 
lots of infrastructure changes).  Currently it is pointing out that while 
`pkg` can return multiple packages for an unversioned package query, the 
backend only returns the first one - I'm not sure as of yet how to proceed 
to resolve this.

There are now two main "helper" subsystems available for the common tasks 
of selecting one package matching a PackageID and either sending the 
package to a function for emitting data from it (GetDetails, GetFiles) or 
adding it to a (currently one package only) job and sending it to a 
function for solving and applying (InstallPackages).  There are a lot of 
optimisations and tweaks that could be made to the query functions, as the 
whole query code has mostly evolved to save duplication rather than being 
designed.

I've also removed most of the dummy boilerplate.  I was keeping this in to 
remind me which features I needed to install, but as the backend is 
becoming more and more useful I thought I'd take it out so no non-GSoC 
code is polluting the main code file.

In all, I thought I was going to be behind my milestones this week, but 
aside from some niggles left to solve later it seems that both local and 
remote package installation are present and informally tested to work in 
certain use cases.

Known bugs:
  - Backend sometimes crashes when moving from SimulateInstallPackages to
    InstallPackages or cancelling; not sure what's causing this at the
    moment, will investigate later.
  - Test for GetDetails expects certain packages to be installed and
    fails when multiple versions of a single package exist, needs some
    thought as to how the backend should solve multiple version problems.
  - Install list produces multiple reinstall/install/update category
    headings due to no sort being produced on the iterator output; may
    fix, may leave as-is.
  - Non-severe errors can cause the installation to fail with a fatal
    error, I need to change the install event handler to stop this.
  - All package arguments must be given as PackageIDs; this is due to
    Resolve not being implemented, may work on this next week.

~ Matt