Hi Guillem, Here is v2 of the series. It no longer contains the --sign-pause fix, since you have merged that locally already. Patch 1/2 is new (the .gitignore change), and patch 2/2 is the updated completion file.
> But it seems some of my changes complement some of yours. So it would > make sense if you adopted some of those into your patch? I fetched your pu/bash-completions branch (4198c5e9b) and compared it with mine. I adopted your approach of completing --sign-command with a curated list: mine used _comp_compgen_commands, which expands to 7024 entries here, so it was not usable interactively. My values differ from yours in two ways. First, neither auto nor none is offered, as dpkg-buildpackage validates the value with find_command() before it reaches the backend (scripts/dpkg-buildpackage.pl:674), so both error out with "sign-command 'auto' not found". Second, the *v commands (sqopv, rsopv, sqv, gpgv-sq and gpgv) are the verification commands (DEFAULT_CMDV), rather than the signing ones. The final list is the DEFAULT_CMD of each supported backend, in the backend preference order from dpkg-buildpackage(1). I did not adopt -x dpkg_architecture, because the xfunc is defined in the dpkg-architecture completion file, which is not in main yet; adding it now makes --host-arch complete nothing. I will switch to it once that file lands. > Right, it looks I was also bothered by this, and ended up adding negated > entries (in a commit in that branch). But I think I do prefer your > proposed fix better. So feel free to provide a patch for that. Patch 1/2 does this. It anchors every entry in scripts/, src/, dselect/ and utils/, which are the four directories containing a completion/bash/ subdirectory. Once it is applied, the negated entries you added in 3c498a0c3 will no longer be needed. > Nice catch as well. I was curious whether the current completion files > were fine, so added this locally now (with attribution), and fixed a > couple of issues with the dpkg-deb completion. Thank you, both for adding the test and for the attribution. Best regards, Cheng Li
