Improve auto-completion of filenames

[benjaminweb]

There are file1.py and file1_test.py only.

hg add file1.py file1…

There, fish could see that file1.py has been provided earlier.
As it makes no sense to repeat the same file name, fish could auto-suggest file1_test.py.

[eush77]

Counter-example:

$ seq 1 5 >one
$ seq 1 2 10 >two
$ cat one two two |sort |uniq -u
2
4

Suppose you have another file two_something in the same directory, then Fish auto-completing two the second time to two_something would make it harder to select just two.

[faho]

Yes, this isn't something that could be enabled for all completions - there would need to be a way to opt-out or opt-in (I think opt-out might be okay for file-completion).

It also can't be done for all kinds of completions in general, since e.g. options might be specified multiple times (rather common for "-v" or "--verbose" options).

Doing it correctly would also be suprisingly hard - since filenames are almost completely unrestricted on UNIXy systems, and since there's quite a variety of cases in which something that looks like a filename may appear in a different context. Long story short: This would pretty much require something like #478.

For suggestions, however, which don't need to be 100% correct - only very likely to be, I think it'd be okay to enable, even without an option to disable.

[pickfire]

Yeah, I would like this features sometimes althought it could cost a bit of performance.

I would be nice to just ignore the repeated pattern for some command so that it doesn't affect mv file.txt file.txt.old.

[benjaminweb]

I specifically address this case: file1.py file1_test.py
+1