Re: Generating manifests for the test runner

On 23/07/14 01:08, Michael[tm] Smith wrote:
> Hi Brad,
> 
> ./tools/scripts/manifest.py --experimental-include-local-changes
> 
> You need to feed the script that
> "--experimental-include-local-changes" switch.

Yes, so fundamentally the idea of the generated manifest file is that
it entirely uses in-test-file data to identify tests and their
properties. Unfortunately this makes it rather slow to regenerate the
manifest for the whole repository since it involves parsing all the
HTML files, among other things. As such we need to support incremental
updates to the manifest. But the combination of incremental updates
between revisions and adding data from uncommitted local files is
slightly non-trivial to support; one doesn't want to end up with items
in the manifest corresponding to files that were never committed and
no longer exist, for example. I am not confident that the current
approach to this in the code is good enough.

It's quite possible that improvements can be made. For example, with
files that have been modified it should first generate the correct
data for HEAD and then separately store the data for the working tree
copy. On update all data for non-comitted modifications would be
discarded. I will be able to look at this relatively soon, I would think.

It also should be possible to ensure that all consumers of the
manifest generate it as needed, so that there is never any need to
manually initiate the generation.

Received on Wednesday, 23 July 2014 08:52:44 UTC