add support for -F/--fixed-strings

You could perhaps turn this on automatically if the pattern is
detected to not contain any metacharacters, which is probably pretty
common.

--
Martin <http://launchpad.net/~mbp/>

On Wed, Mar 17, 2010 at 5:45 AM, Martin Pool <email address hidden> wrote:
> You could perhaps turn this on automatically if the pattern is
> detected to not contain any metacharacters, which is probably pretty
> common.
>

Sounds like a good idea.

Another option could be that, rather than doing magic, we could
make this the default and have a -e flag for taking this as pattern.

As per the grep manual, -F will:

Interpret PATTERN as a list of fixed strings, separated by
newlines, any of which is to be matched. (-F is specified by
POSIX.)

I am wonder how important the "list" part of "list of fixed strings"
is. I think supporting just one string would be very fast and cover
most cases.

What do you think?

On 17 March 2010 13:01, Parth Malwankar <email address hidden> wrote:
> On Wed, Mar 17, 2010 at 5:45 AM, Martin Pool <email address hidden> wrote:
>> You could perhaps turn this on automatically if the pattern is
>> detected to not contain any metacharacters, which is probably pretty
>> common.
>>
>
> Sounds like a good idea.
>
> Another option could be that, rather than doing magic, we could
> make this the default and have a -e flag for taking this as pattern.

I think people would be confused if they give a regexp and it's not
matched as such.

>
> As per the grep manual, -F will:
>
> Interpret  PATTERN  as a list of fixed strings, separated by
> newlines, any of which is to be matched.  (-F is specified by
> POSIX.)
>
> I am wonder how important the "list" part of "list of fixed strings"
> is. I think supporting just one string would be very fast and cover
> most cases.

I think so too. I've never used that grep feature.

--
Martin <http://launchpad.net/~mbp/>

On Wed, Mar 17, 2010 at 7:46 AM, Martin Pool <email address hidden> wrote:
> On 17 March 2010 13:01, Parth Malwankar <email address hidden> wrote:
>> On Wed, Mar 17, 2010 at 5:45 AM, Martin Pool <email address hidden> wrote:
>>> You could perhaps turn this on automatically if the pattern is
>>> detected to not contain any metacharacters, which is probably pretty
>>> common.
>>>
>>
>> Sounds like a good idea.
>>
>> Another option could be that, rather than doing magic, we could
>> make this the default and have a -e flag for taking this as pattern.
>
> I think people would be confused if they give a regexp and it's not
> matched as such.
>

Sounds reasonable, so if the pattern isalnum() == True, I will go
with -F (apart from the explicit -F). That should be a safe enough
whitelist.

Added -F/--fixed-string support. This is different from --fixed-strings in that the PATTERN is assumed to be a single string (and not newline delimited strings). As an optimization, simple aphanumeric PATTERN implicitly set -F.

For a large tree like emacs, working copy search time has gone from ~4.5s to ~2.4s on my system.