Re: [RFC] Cleanup for make_source_files_completion_list

[Keith Seitz <keiths@redhat.com>]

On 05/09/2013 04:45 PM, Doug Evans wrote:
> The caller of make_source_files_completion_list explains the decision
> to pass "text" for "word" here:
>
>        /* If we only have file names as possible completion, we should
>           bring them in sync with what rl_complete expects.  The
>           problem is that if the user types "break /foo/b TAB", and the
>           possible completions are "/foo/bar" and "/foo/baz"
>           rl_complete expects us to return "bar" and "baz", without the
>           leading directories, as possible completions, because `word'
>           starts at the "b".  But we ignore the value of `word' when we
>           call make_source_files_completion_list above (because that
>           would not DTRT when the completion results in both symbols
>           and file names), so make_source_files_completion_list returns
>           the full "/foo/bar" and "/foo/baz" strings.  This produces
>           wrong results when, e.g., there's only one possible
>           completion, because rl_complete will prepend "/foo/" to each
>           candidate completion.  The loop below removes that leading
>           part.  */
>
> Note that if you remove "word" from make_source_files_completion_list
> then add_filename_to_list collapses to a trivial function (which would
> otherwise be great except there's basic core functionality that I
> think should be kept).  :-)

Indeed. I looked into this a bit further after Tom's comments, too. As 
it is, I am convinced that this patch should not be accepted. [aka, Tom 
is right -- those callers of make_source_files_completion_list which 
pass text twice are suspicious.]

However, I think that the comment quoted above (and the surrounding 
block of code which does the copying) should be removed and the callers 
of make_source_files_completion list should pass text AND word instead 
of text and text. add_filename_to_list (in symtab.c) does this text - 
word adjustment already, so it is not necessary to repeat it here. WDYT?

> OTOH the "text" arg to completion_list_add_name can go, it's unused.

I'll take a look at that, too, while I'm at it. That didn't come up 
during my completer implementation, so it escaped scrutiny.

Keith