Enhancements to bash completions for better file/directory handling (take 2) #782
Conversation
Allows parameter types to implement their own completions.
The -o default will cause the shell to complete paths in any case that the program did not provide completions in COMPREPLY, this is undesirable as it will result in completing filenames when the next argument or option in context is not a filename. We use nospace here so we can control whether to append a space or not to the completions, this will allow us to complete paths with python code and allow completing more dynamically (so that after completing a subdirectory there is no space appended and one can continue to complete the next subdir).
…tions Instead of special casing the Choice type. Note that this checks the user provided 'autocompletions' list in advance of consulting the parameter class, this should allow users to override the default completion list of a parameter. Some minor adjustments made to append a space to builtin completion lists of commands and option names so that this works seamlessly with the new "complete -o nospace" approach.
Slightly changed semantic for the autocompletion function which can be specified in a decorator, now point out the semantic difference of appending a space to a returned completion suggestion or not. This should not be an API break since the autocompletion function was added recently and has never been in a stable/official release.
After changing the bash wrapping script to invoke complete with the nospace option, the decision to append a space and decide to start completing the next program argument falls on click. So now all results from get_choices() which are entirely "complete" and cannot be further completed have an appended space. Updated test case to expect this behavior from get_choices().
Since we access the filesystem for tab completion of files and directories, we should be defensive and just ignore any errors which might occur due to permission or other errors.
This was broken due to the way we split the input incomplete path on the os separator token, when there is nothing to the left of the last filesystem separator, we know we are trying to complete from the root. For windows, this should not happen since there will always be a drive letter on the left of a valid terminating path separator.
We only want to filter out non-directories when searching for a directory. We still want directory completions when searching for a file.
Ensure we still complete directory names when searching for a regular file, we only want to disable non directory results when searching for a directory path.
|
Addressed some corner cases regarding path name completions in some follow up commits instead of squashing those and rebasing the branch, as that seems to cause a merge request to be automatically closed. I can rebase the branch locally so that these changes are squashed into the one commit implementing completions for file & directory paths, if that is desirable. |
|
Completion has changed significantly since this PR was posted. Would you or @stopthatcow review this to see if it's still needed and can be rebased? |
|
There are definitely some attractive aspects of this PR. I like the idea of delegating some completion to the type, and path completion would be very nice to have. I can look into rebasing it on top of master this weekend, unless of course @gtristan wants to take a stab at it himself! |
It's great to see some activity on this front ! I would certainly love to upstream this stuff, as we're doing some pretty ugly hacks which reimplement the completions mechanism in order to get these features in BuildStream. The most important part for us is the ability to perform partial completions, since we do completions of filenames which are intentionally not in the As far as I know, this can only be done by enabling I'll have to refresh my memory on how compatible my proposed changes were, but I'm pretty confident that by delegating completions to the type and to delegated user functions; we can make sure that the behavior of allowing partial completions is opt in. |
|
@davidism Thanks for reviewing and for everyone who picked up my abandoned patches :) I will try to find time to upgrade click in BuildStream and drop our custom code in favor of this :) |
This addresses the issues outlined in #780
completions()method for BoolParamType, File and Path parameter types tooThis behaves all around much better:
Note: This is take two of pull request #781 ... I rebased my branch to have a more coherent history and did not expect github to delete the PR as a result, sorry.