Re: [PATCH/RFC] blame: accept multiple -L ranges

Eric Sunshine [off-list ref] writes:

As implemented by this patch, the behavior of git-blame with multiple
-L's is consistent with that of git-log with multiple -L's. The
implemented behavior feels intuitive to me, but I can see how the
behavior you suggest could feel intuitive to others.

If I re-do the patch to work the way you describe above, how should we
deal with the inconsistent behaviors between the two commands?

To be extremely honest, I do not care too deeply about what log -L
does today, because it is still in "may have rough edges but is an
interesting toy to play with" state in my mind ;-)

The suggestion to error out was more about "start simple, strict and
obvious to make it easy to explain" and nothing else.  If we start
with a simple and strict version, we can later loosen it without
making an input that was valid earlier invalid.  If we start with
too loose, on the other hand, it would be hard to tighten it later.

But the only two things I care deeply about are, in a file whose
contents is:

	C
        B
        A
        B
        C
        D

 (1) The range "-L /A/,/B/" finds the first A from the beginning,
     and then chooses B that comes _after_ it, making it equivalent
     to -L3,4 (not -L3,2 or -L2,3).

 (2) In the ranges "-L <anything>,/B/ -L /C/,<anything>", the
     beginning of the second range is found by choosing C that comes
     _after_ the end of the previous range (/B/ may choose either
     the second or the 4th line, and the only C that comes after
     either of them is the 5th line and that is where the second
     range should begin, not at the beginning of the file).  The
     same for "-L 1,3 -L /C/" (only C that comes after 3 is eligible
     to be the beginning of the second range).

I view it as a nice addition to coalesce two overlapping ranges
given exactly by numbers, e.g. "-L 100,200 -L 50,102".  I do not
have a strong objection to it, as long as it does not interfere
negatively with ranges specified by patterns.

Thanks.

On Tue, Jul 9, 2013 at 12:39 PM, Junio C Hamano [off-list ref] wrote:

Eric Sunshine [off-list ref] writes:

quoted

As implemented by this patch, the behavior of git-blame with multiple
-L's is consistent with that of git-log with multiple -L's. The
implemented behavior feels intuitive to me, but I can see how the
behavior you suggest could feel intuitive to others.

If I re-do the patch to work the way you describe above, how should we
deal with the inconsistent behaviors between the two commands?

The suggestion to error out was more about "start simple, strict and
obvious to make it easy to explain" and nothing else.  If we start
with a simple and strict version, we can later loosen it without
making an input that was valid earlier invalid.  If we start with
too loose, on the other hand, it would be hard to tighten it later.

Makes sense.

But the only two things I care deeply about are, in a file whose
contents is:

        C
        B
        A
        B
        C
        D

 (1) The range "-L /A/,/B/" finds the first A from the beginning,
     and then chooses B that comes _after_ it, making it equivalent
     to -L3,4 (not -L3,2 or -L2,3).

 (2) In the ranges "-L <anything>,/B/ -L /C/,<anything>", the
     beginning of the second range is found by choosing C that comes
     _after_ the end of the previous range (/B/ may choose either
     the second or the 4th line, and the only C that comes after
     either of them is the 5th line and that is where the second
     range should begin, not at the beginning of the file).  The
     same for "-L 1,3 -L /C/" (only C that comes after 3 is eligible
     to be the beginning of the second range).

Thinking aloud: Thus, for "-L 1,5 -L /C/", no C would be found.

I view it as a nice addition to coalesce two overlapping ranges
given exactly by numbers, e.g. "-L 100,200 -L 50,102".  I do not
have a strong objection to it, as long as it does not interfere
negatively with ranges specified by patterns.

Okay, I will base my rewrite upon the above constraints, along with
other observations from your initial response.

-- ES