ediff refinement issues

ediff refinement issues

[Samuel Wales]

Ediff lets you show individual words that are different, and
it also shows you when there are only whitespace differences
in paragraphs.

I am having issues with each of these two related features.

  1) When there are only whitespace differences in normal
     paragraphs, such as by refilling, ediff works well.  It
     says there are only whitespace differences and does not
     highlight any words.

     However, if the paragraphs are commented (for example,
     with ;;; in elisp or # in shell), it highlights the ;;;
     or the #.  it also sometimes highlights words as a side
     effect.

     I want ediff to show refilled commented paragraphs as
     only having whitespace differences.

  2) When I change a date like 2011-01-01, ediff highlights
     the entire date even if I only changed one part of it.

     For example, if I change 01 to 02, it will still
     highlight the entire date.

     I want it to show only the part I changed.


There are variables ediff-word-1 ediff-word-2 ediff-word-3
ediff-word-4 that are supposed to customize this.  I tried
them in various ways and they didn't produce the results I
wanted.

How do you use those variables to do this?

Thanks.

RE: ediff refinement issues

[Drew Adams]

>   1) When there are only whitespace differences in normal
>      paragraphs, such as by refilling, ediff works well.  It
>      says there are only whitespace differences and does not
>      highlight any words.
> 
>      However, if the paragraphs are commented (for example,
>      with ;;; in elisp or # in shell), it highlights the ;;;
>      or the #.  it also sometimes highlights words as a side
>      effect.

I don't see that.  Maybe give a concrete example.  What I see is that whitespace
is either ignored everywhere or it is not ignored at all (toggle this with
`##').

>      I want ediff to show refilled commented paragraphs as
>      only having whitespace differences.

Are you sure that the paragraphs themselves have only whitespace differences?
You say that `;;;' is highlighted as different.  Do you mean that the `;;;' is
highlighted in both of these identical lines?

;;; commented line

;;; commented line

>   2) When I change a date like 2011-01-01, ediff highlights
>      the entire date even if I only changed one part of it.
>      For example, if I change 01 to 02, it will still
>      highlight the entire date.
>      I want it to show only the part I changed.
> 
> There are variables ediff-word-1 ediff-word-2 ediff-word-3
> ediff-word-4 that are supposed to customize this.  I tried
> them in various ways and they didn't produce the results I
> wanted.  How do you use those variables to do this?

I've never played with that, and I don't know of any command or option that
helps here, but maybe someone else can help.  Looking in the code a bit, I see
this:

1. Those are defvars, not defcustoms.
2. They are buffer-local.

So I tried this in each of two buffers that had only this, respectively:
"2011-01-01" and "2011-01-02": `M-: (setq ediff-word-1 "[:word]")'.

Then, hitting `!' showed a refinement that distinguished "2011" from the rest.
I would have expected the "01" / "02" difference to be distinguished. Maybe you
can play around a little more this way to get what you want.

Re: ediff refinement issues

[Samuel Wales]

On 2011-11-26, Drew Adams <drew.adams@oracle.com> wrote:
> Are you sure that the paragraphs themselves have only whitespace
> differences?

Yes of course.  I want ^;;; or ^# to act as whitespace in this case.

It is possible that those variables (which are in the Customization
section of the ediff manual) are insufficient to do what I want here.
It is possible that I need a new ediff-forward-word-function (which is
also in that section).

One issue is that I want only ^;;; or ^# to be treated as whitespace.
Not ;;; or # in the middle of text.

It is really simple.  I don't see why I should have to look at
highlighted spurious things when all I did was fill a paragraph.

> You say that `;;;' is highlighted as different.  Do you mean that the `;;;'
> is
> highlighted in both of these identical lines?
>
> ;;; commented line
>
> ;;; commented line

No, because they are identical.  But if you had an unfilled paragraph
there instead of a single line, and ediffed the filled and unfilled
versions, you might understand.

> 2. They are buffer-local.

What ediff hook works best here?

RE: ediff refinement issues

[Drew Adams]

> Yes of course.  I want ^;;; or ^# to act as whitespace in this case.

AFAIK, there is no way to tell ediff to ignore differences in text that matches
an arbitrary regexp.  IOW, it ignores only whitespace when you tell it to ignore
whitespace. ;-)

Ignoring whitespace is a built-in feature of ediff.  IIUC, what you are asking
for is not a feature provided by ediff.

I have long wanted the ability to tell ediff to ignore (i.e., consider as the
same) all matches for a given pattern.  AFAIK, that feature is waiting to be
implemented. ;-)

> One issue is that I want only ^;;; or ^# to be treated as whitespace.
> Not ;;; or # in the middle of text.

Neither case will ever be treated as whitespace.  Such patterns do not represent
whitespace.

> It is really simple.  I don't see why I should have to look at
> highlighted spurious things when all I did was fill a paragraph.

Filling a paragraph typically changes more than whitespace.

;;; aaa bbb ccc ddd
;;; eee

and 

;;; aaa bbb ccc
;;; ddd eee

wrt more than whitespace.  Ediff shows you the difference.
`;;;' is not whitespace.

> But if you had an unfilled paragraph... and ediffed the
> filled and unfilled versions, you might understand.

They differ wrt non-whitespace text, namely, the comment chars.

AFAICT, in what you describe:

1. Ediff is acting as advertised.
2. There is no option/command to have ediff ignore differences in text that
matches a pattern other than whitespace.

Again, maybe someone else can help more.

Re: ediff refinement issues

[Thien-Thi Nguyen]

() "Drew Adams" <drew.adams@oracle.com>
() Sat, 26 Nov 2011 10:38:28 -0800

   2. There is no option/command to have ediff ignore differences in
      text that matches a pattern other than whitespace.

If Ediff usees GNU diff, perhaps there is a way to specify ‘-I REGEXP’.

Re: ediff refinement issues

[Samuel Wales]

On 2011-11-29, Thien-Thi Nguyen <ttn@gnuvola.org> wrote:
>    2. There is no option/command to have ediff ignore differences in
>       text that matches a pattern other than whitespace.
>
> If Ediff usees GNU diff, perhaps there is a way to specify ‘-I REGEXP’.

Hi TTN,

I think that would ignore comments entirely, instead of ignoring
formatting in comments.  No?

Samuel

Re: ediff refinement issues

[Thien-Thi Nguyen]

() Samuel Wales <samologist@gmail.com>
() Tue, 29 Nov 2011 10:27:54 -0700

   On 2011-11-29, Thien-Thi Nguyen <ttn@gnuvola.org> wrote:
   >    2. There is no option/command to have ediff ignore differences in
   >       text that matches a pattern other than whitespace.
   >
   > If Ediff usees GNU diff, perhaps there is a way to specify ‘-I REGEXP’.

   I think that would ignore comments entirely, instead of ignoring
   formatting in comments.  No?

You are right.  It looks like ‘-I REGEXP’ is not what you seek.

Re: ediff refinement issues

[Samuel Wales]

On 11/26/11, Drew Adams <drew.adams@oracle.com> wrote:
>>   1) When there are only whitespace differences in normal
>>      paragraphs, such as by refilling, ediff works well.  It
>>      says there are only whitespace differences and does not
>>      highlight any words.
>>
>>      However, if the paragraphs are commented (for example,
>>      with ;;; in elisp or # in shell), it highlights the ;;;
>>      or the #.  it also sometimes highlights words as a side
>>      effect.
>
> I don't see that.  Maybe give a concrete example.  What I see is that
> whitespace
> is either ignored everywhere or it is not ignored at all (toggle this with
> `##').

you are confusing different features.  take a commented paragraph with
a few long lines, save it to a.  fill it.  save it to b.  now do diff
mode on it and also ediff on it.

you should notice that diff mode correctly notices tht the only
non-whitgespaceish ifference is the comment prefix.  ediff thinks
words changed.

Re: ediff refinement issues

[Samuel Wales]

for purpoes of the test do something like this:

# asdfjkansdkljfan ksd fnak nsdflkadsnfkansdkfnakdsnf
# aklsfdnakljfnsdkasdnfja sdn fkljad nsfkj ansdkfn aksdjfna
# ksjfndk ajdnsf kajsndkfjansdkfjn ja nsdkjfn askdj fnakjs fdnakj
# nsdfkaj dns

then fill to a narrow fill column.

On 3/28/21, Samuel Wales <samologist@gmail.com> wrote:
> On 11/26/11, Drew Adams <drew.adams@oracle.com> wrote:
>>>   1) When there are only whitespace differences in normal
>>>      paragraphs, such as by refilling, ediff works well.  It
>>>      says there are only whitespace differences and does not
>>>      highlight any words.
>>>
>>>      However, if the paragraphs are commented (for example,
>>>      with ;;; in elisp or # in shell), it highlights the ;;;
>>>      or the #.  it also sometimes highlights words as a side
>>>      effect.
>>
>> I don't see that.  Maybe give a concrete example.  What I see is that
>> whitespace
>> is either ignored everywhere or it is not ignored at all (toggle this
>> with
>> `##').
>
> you are confusing different features.  take a commented paragraph with
> a few long lines, save it to a.  fill it.  save it to b.  now do diff
> mode on it and also ediff on it.
>
> you should notice that diff mode correctly notices tht the only
> non-whitgespaceish ifference is the comment prefix.  ediff thinks
> words changed.
>


-- 
The Kafka Pandemic

Please learn what misopathy is.
https://thekafkapandemic.blogspot.com/2013/10/why-some-diseases-are-wronged.html

Re: ediff refinement issues

[Eli Zaretskii]

> From: Samuel Wales <samologist@gmail.com>
> Date: Sun, 28 Mar 2021 21:48:18 -0700
> Cc: help-gnu-emacs@gnu.org
> 
> for purpoes of the test do something like this:
> 
> # asdfjkansdkljfan ksd fnak nsdflkadsnfkansdkfnakdsnf
> # aklsfdnakljfnsdkasdnfja sdn fkljad nsfkj ansdkfn aksdjfna
> # ksjfndk ajdnsf kajsndkfjansdkfjn ja nsdkjfn askdj fnakjs fdnakj
> # nsdfkaj dns
> 
> then fill to a narrow fill column.

Please submit a full bug report with the details, TIA.

Re: ediff refinement issues

[Samuel Wales]

done [as best as poss], thanks.

On 3/28/21, Eli Zaretskii <eliz@gnu.org> wrote:
> Please submit a full bug report with the details, TIA.

Re: ediff refinement issues

[Stefan Monnier]

>>>      However, if the paragraphs are commented (for example,
>>>      with ;;; in elisp or # in shell), it highlights the ;;;
>>>      or the #.  it also sometimes highlights words as a side
>>>      effect.

FWIW, this is largely unavoidable.
Refilling has an effect similar to replacing

    foo bar ;; baz

with

    foo ;; bar baz

So the "fine" diff indicates that semi colons were removed at one place
and inserted at the other.  In order to "abstract away" semi-colons
which are placed at the beginning of all lines, it would take extra work
and neither ediff nor smerge does that.  I haven't thought about how
easy/hard it might be to implement such a thing (but at first glance,
I'd say it's probably not trivial).

As for why ediff sometimes highlights words and not smerge, I suspect
it's mostly an accident (linked to the details of how the fine-diffs are
defined in the two cases).  E.g. in the above example you could also say
that what happened is that "bar" was removed at one place and inserted
at another.  That's also valid, tho less good.  Maybe Ediff gets it
"less good" more often than Smerge because it doesn't have the
equivalent of `smerge-refine-weight-hack` which marks "move ;;" as less
costly than "move bar" in the above example?


        Stefan

Re: ediff refinement issues

[Michael Heerdegen]

"Drew Adams" <drew.adams@oracle.com> writes:

> > There are variables ediff-word-1 ediff-word-2 ediff-word-3
> > ediff-word-4 that are supposed to customize this.  I tried
> > them in various ways and they didn't produce the results I
> > wanted.  How do you use those variables to do this?
>
> I've never played with that, and I don't know of any command or option
> that helps here, but maybe someone else can help.  Looking in the code
> a bit, I see this:
>
> 1. Those are defvars, not defcustoms.
> 2. They are buffer-local.
>
> So I tried this in each of two buffers that had only this, respectively:
> "2011-01-01" and "2011-01-02": `M-: (setq ediff-word-1 "[:word]")'.

Yes.  And I think one should better add the hyphen character back to one
of the other word class characters.

> Then, hitting `!' showed a refinement that distinguished "2011" from
> the rest.  I would have expected the "01" / "02" difference to be
> distinguished.

That can happen when diff or word counting is wrong.  As you said: these
variables are buffer local, and Ediff uses a separate buffer to
"wordify" for the fine diffs, and in that buffer the original bindings
of ediff-word-* are visible, so you get an inconsistency in the
definition of word classes.

In summary, this should work ok:

  ;; e-w-1 minus _ and -
  (setq-default ediff-word-1 "[:word:]")
  ;; e-w-3 plus _  and -
  (setq-default ediff-word-3 "`'?!:;\"{}[]()_-")

Michael.

Re: ediff refinement issues

[Peter Münster]

On Sat, Nov 26 2011, Samuel Wales wrote:

>      I want ediff to show refilled commented paragraphs as
>      only having whitespace differences.

You could implement this feature. But then, be careful with
commented code:

file A:
shell-command

file B:
# shell-command

"^#" should not be ignored here.

-- 
           Peter