On 4/12/20 11:14 AM, Daniel Albers wrote:

This saves users and gitlab.com a 2.2 GB fetch, which, aside from resources, takes a significant amount of time.

The $SRCDEST variable means this will only occur once per user. Future rebuilds will only git fetch.

Your PKGBUILD deletes the clone every time src/ is deleted, e.g. by makepkg -C, which means for users who did use $SRCDEST they waste far, far more resources, both theirs and gitlab's, and spend longer doing it. That is, in the long run.

Users whose AUR helper clones into /tmp and automatically deletes the clone afterward, are being generally wasteful and this bites them in the case of large sources like inkscape. Because they are doing it wrong, a special PKGBUILD doing shallow clones does reduce their bandwidth consumption, but that isn't a reason to violate the AUR hosting policy.

Is there a supported way to use shallow clones?

Speaking with my pacman developer hat on: No, we have rejected this feature request numerous times. Your pkgver() function is broken because pkgver functions depend on the full git clone, and yours is instead reporting falsehoods.

It's a textbook example of why it is forbidden in the first place.