Files from recursive file search should always be sorted

[rillig]

When I run ack return src, the files from the src directory are listed in apparently random order. This order can be changed by passing the --sort-files option. But then, all files are sorted, no matter in which order they were given on the command line.

• When scanning a directory recursively for files, these files should always be sorted alphabetically.
• The option --sort-files should only affect the files that were given explicitly on the command line.

Example: When running ack searchTerm /usr/include include src checks, the files should be displayed in the following order:

1. all files from /usr/include, sorted alphabetically
2. all files from include, sorted alphabetically
3. all files from src, sorted alphabetically
4. all files from checks, sorted alphabetically

[hoelzro]

@rillig The way ack find files to search (regarding order, and ignoring things like filters) is this:

Get list of files from current directory
If --sort-files is on, sort that list in lexicographical order
For each file in that list:
    If the file is a directory, recurse into it
    Otherwise, search its contents

I don't think @petdance will want to change this behavior, mainly since it would affect and surprise all of the users who understand ack to behave as specified above.

[petdance]

I don't know. It's an interesting idea. I can see why you'd want to specify a certain order for things on the command line.

[rillig]

To prevent a misunderstanding:

• Currently, when getting the list of files from a directory, the files are not sorted at all (according to @hoelzro’s comment).
• I want the files from a single directory to be sorted alphabetically, since “filesystem read order” is mostly arbitrary and not useful.
• I don’t want any other behavior to be changed.

[petdance]

You can get the files sorted with --sort-files, but then you'll also get your starting directories sorted as well.

[epa]

Sorting the files is probably more useful than the current behaviour, but perhaps marginally slower (on rotating storage at least). However it might not be good to commit to sorted filenames as documented behaviour, since it precludes optimizations like scanning the tiny files first and leaving huge ones for later. I guess --sort-files-within-directory as an explicit option would work.