[Gnu-arch-users] Out-of-order patch merging and tla update

[Martin Langhoff]

Using Arch for a web-based app (Moodle) I've found a bit of a weird
situation I expected tla to deal with more gracefully.

We deploy to production doing `tla update`, but sometimes, we deploy
individual patches out-of-order using get-patch and do-changeset. This
happens because patches are quite self-contained, and some may be
getting extra testing or may require off-peak-hour deployment.

Whenever the working copy has patches deployed out-of-order, `tla
changes` was returning a lot of funny stuff. Obviously, the revision
library had been built with the latest complete version, and was
showing "reverse patches" for all the few patches we had skipped. I
didn't worry.

When I did `tla update`, however, things were a bit funny, and the
regular ritual of 'undo-merge-redo' went seriously wrong. Valiantly, I
resisted hitting ctrl-c to stop the mess I had set in motion. After it
was done, I did doing 'tla undo' afterwards, to ensure I had a clean
patchlevel. It means, however, that any local changes we had are
buried in the latest undo directory, together with a lot of noise.

Of course, the right thing to do would have been to deploy the patches
we had skipped until I had a clean patchlevel, and only do tla update
there.

Are there ways to achieve the same goals without tla getting so
confused? A "production" branch were we merge things to?

cheers,


martin