[Geany-Devel] Improving FiF

[Matthew Brush]

On 13-10-09 07:15 PM, Lex Trotman wrote:
> [...]
>
>> 3. For non-recursive searches, grep does not allow a directory to be
>>> specified instead of file. To work around that, we read the directory,
>>> back-parse all --include=*.x patterns, and match them manually.
>>>
>>> Proposition: grep -rl --include=*.c --exclude-dir=[^.]*
>>> --exclude-dir=.?* void . :)
>>>
>>> We pass our Directory as a workind  directory to spawn, and the
>>> recursive search already uses . as a grep FILE argument.
>>> --exclude-dir is supported since more than 5 years.
>>>
>>>
>> Is there any reason we cannot just walk the directory/subdirectories
>> ourselves and search the files using GRegex stuff? The pattern syntax would
>> then be the exact same as the other search dialogs and it would save having
>> to not only spawn a subprocess but also remove the dependency on grep (only
>> a problem on Windows). Also, as a future optimization, if any of the files
>> to search are open in Geany, we could search their document buffer directly
>> from memory rather than having to do any file IO for them. GIO/GFile has
>> all the stuff needed to walk a directory tree and open files both
>> asynchronously and portably IIRC.
>>
>
> *Technically* the only issue I know of is that last time I looked PCRE was
> much less optimised than grep (partly because PCRE is more general) so
> searching a big tree of big files may be noticeably slower.
>
> But this is more code that needs to be written, debugged and maintained.
>   Pull requests (with an ongoing support pledge) are welcome.
>

I'll gladly write and help maintain it if I don't have to bikeshed over 
using stupid old versions of GIO which might not have the features 
needed[1] to implement it properly without dirty hacks or #ifdef stuff :)

> The benefits of control and consistency (and as you pointed out on IRC to
> search the buffers not the files for open documents) may not outweigh the
> cost of doing it.
>

Cheers,
Matthew Brush

[1]: I haven't investigated deeply, the docs don't list "since" versions 
for some of the functions that would be used.