transpose-sexps

transpose-sexps

[Paul Pogonyshev]

Put this in *scratch*, put point over `f' letter and hit C-M-t:

	"t" "f"

I get this:

	"tf" ""

which I consider utterly improper.  Moreover, it differs from what
I get by M-t.

Given the recent bug I discovered in Python mode, looks like sexp
scanning has been broken.

Paul

Re: transpose-sexps

[Stuart D. Herring]

> Put this in *scratch*, put point over `f' letter and hit C-M-t:
>
> 	"t" "f"
>
> I get this:
>
> 	"tf" ""
>
> which I consider utterly improper.  Moreover, it differs from what
> I get by M-t.

See the doc:

Like M-t but applies to sexps.
Does not work on a sexp that point is in the middle of
if it is a list or string.

It assumes that you are outside of strings or lists, so it saw
             " "
and
              f
as your two sexps, and exchanged them.  It's just confusing that it looks
like the f was moved into the other string.

Davis

-- 
This product is sold by volume, not by mass.  If it appears too dense or
too sparse, it is because mass-energy conversion has occurred during
shipping.

Re: transpose-sexps

[Paul Pogonyshev]

Stuart D. Herring wrote:
> It assumes that you are outside of strings or lists, so it saw
>              " "
> and
>               f
> as your two sexps, and exchanged them.  It's just confusing that it looks
> like the f was moved into the other string.

It's not just confusing, it's a bug.  If it doesn't work and this is not
(easily) fixable, make it do nothing or report an error.  Producing confusing
results is worse than doing nothing.

Paul

Re: transpose-sexps

[Stuart D. Herring]

> It's not just confusing, it's a bug.  If it doesn't work and this is not
> (easily) fixable, make it do nothing or report an error.  Producing
> confusing
> results is worse than doing nothing.
>
> Paul

In the list case, it's not a bug, because it's impossible to guess the
user's intentions.  In
  (-!-foo bar) (baz spam)
with point as always at -!-, is the "current sexp" (to be transposed with
the following one) the symbol foo, or the list (foo bar)?

For the string case, it would be possible to "fix" this, but to do nothing
or report an error requires detecting the case just as much as doing the
right thing.  It is the detection which is hard -- just ask the CC mode
people trying to sort out defun detection.  Even in Lisp it's equivalent
to asking how many unescaped " characters precede point, not counting
those in comments (whose identification itself requires counting " in the
same manner).  Perhaps some clever caching or hints from the user could
make that reasonable to implement, but there is no useful change at all
that is easily made.

Davis

-- 
This product is sold by volume, not by mass.  If it appears too dense or
too sparse, it is because mass-energy conversion has occurred during
shipping.

Re: transpose-sexps

[Andreas Schwab]

Paul Pogonyshev <pogonyshev@gmx.net> writes:

> Stuart D. Herring wrote:
>> It assumes that you are outside of strings or lists, so it saw
>>              " "
>> and
>>               f
>> as your two sexps, and exchanged them.  It's just confusing that it looks
>> like the f was moved into the other string.
>
> It's not just confusing, it's a bug.  If it doesn't work and this is not
> (easily) fixable, make it do nothing or report an error.  Producing confusing
> results is worse than doing nothing.

How can you know that " " and f aren't actually the sexps that are
supposed to be transposed?

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: transpose-sexps

[Paul Pogonyshev]

Andreas Schwab wrote:
> Paul Pogonyshev <pogonyshev@gmx.net> writes:
> 
> > Stuart D. Herring wrote:
> >> It assumes that you are outside of strings or lists, so it saw
> >>              " "
> >> and
> >>               f
> >> as your two sexps, and exchanged them.  It's just confusing that it looks
> >> like the f was moved into the other string.
> >
> > It's not just confusing, it's a bug.  If it doesn't work and this is not
> > (easily) fixable, make it do nothing or report an error.  Producing confusing
> > results is worse than doing nothing.
> 
> How can you know that " " and f aren't actually the sexps that are
> supposed to be transposed?

As long as other other "sexp" is empty it doesn't make sense to
"transpose" with it.  Humans are not computers, they will get only
confused by it.  Or what do you mean by " " sexp?

Paul

Re: transpose-sexps

[Richard Stallman]

    It's not just confusing, it's a bug.

The Emacs sexp commands assume that point is not inside a string.
That is not a bug.

It might now be possible to use syntax-ppss to make it smarter.
We should not think about that now, however.

Re: transpose-sexps

[Miles Bader]

Paul Pogonyshev <pogonyshev@gmx.net> writes:
> As long as other other "sexp" is empty it doesn't make sense to
> "transpose" with it.  Humans are not computers, they will get only
> confused by it.  Or what do you mean by " " sexp?

A string is an sexp.  " " is a string containing a single space.

Emacs sees on the left side of point, a sexp: " ", and on the right side
of point, another sexp: f.  It swaps them.  That's what the command
does.

[Maybe, as Richard mentioned, it should be more friendly by detecting
that it's inside a string and do... something... in that case (what??),
but detecting that is not particularly trivial (and so is not a suitable
change just before a release).]

-Miles
-- 
Come now, if we were really planning to harm you, would we be waiting here,
 beside the path, in the very darkest part of the forest?

Re: transpose-sexps

[Andreas Schwab]

Paul Pogonyshev <pogonyshev@gmx.net> writes:

> As long as other other "sexp" is empty

I don't see any empty sexp here.

> it doesn't make sense to
> "transpose" with it.  Humans are not computers, they will get only
> confused by it.  Or what do you mean by " " sexp?

See Stuart's mail.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: transpose-sexps

[Paul Pogonyshev]

Miles Bader wrote:
> Paul Pogonyshev <pogonyshev@gmx.net> writes:
> > As long as other other "sexp" is empty it doesn't make sense to
> > "transpose" with it.  Humans are not computers, they will get only
> > confused by it.  Or what do you mean by " " sexp?
> 
> A string is an sexp.  " " is a string containing a single space.
> 
> Emacs sees on the left side of point, a sexp: " ", and on the right side
> of point, another sexp: f.  It swaps them.  That's what the command
> does.

I see now.

> [Maybe, as Richard mentioned, it should be more friendly by detecting
> that it's inside a string and do... something... in that case (what??),
> but detecting that is not particularly trivial (and so is not a suitable
> change just before a release).]

It is probably possible to ask current major mode if we are in a string.
If it is not possible, than it is probably worth it to invent a consistent
interface that major modes would implement.  I agree that it is not to be
done now, of course.  (And while I'm at it, it's a pity that "now" spans
the last two years or so.)

Paul

Re: transpose-sexps

[Stuart D. Herring]

> [Maybe, as Richard mentioned, it should be more friendly by detecting
> that it's inside a string and do... something... in that case (what??),
> but detecting that is not particularly trivial (and so is not a suitable
> change just before a release).]

I can answer the "what??": the same thing it does when it's inside a sexp
that's a symbol: transpose the string/symbol containing point and the sexp
following that.  This is the convention also followed by M-t.

Sometimes confusion arises because [C-]M-t at the beginning of a word
[sexp] exchanges it with the previous one, so that for these commands an
object includes the point position that begins the next object. 
`transpose-sentences' behaves similarly.

More confusion: C-x C-t trades current and previous line whether or not
point is at the beginning of a line (which is inconsistent either in
direction or in the defined extent of each line), and
`transpose-paragraphs' trades current paragraph with the next one even if
point is just at its beginning, but trades previous and next if point is
between two.

Davis

-- 
This product is sold by volume, not by mass.  If it appears too dense or
too sparse, it is because mass-energy conversion has occurred during
shipping.

Re: transpose-sexps

[Stuart D. Herring]

> It is probably possible to ask current major mode if we are in a string.
> If it is not possible, than it is probably worth it to invent a consistent
> interface that major modes would implement.  I agree that it is not to be
> done now, of course.  (And while I'm at it, it's a pity that "now" spans
> the last two years or so.)

Again, this is not a question of inventiveness, consistency, or any other
kind of design or programming at all.  It's a mathematical question,
really: it is easy to ask the question and to devise a mechanism for doing
so, but it is algorithmically non-trivial to determine the answer.  As
Richard has said, it might be possible to use the (new, complicated)
syntax caching support to answer this question in a reasonable amount of
time, at which point we can consider making C-M-t consult that.

Davis

-- 
This product is sold by volume, not by mass.  If it appears too dense or
too sparse, it is because mass-energy conversion has occurred during
shipping.

Re: transpose-sexps

[Paul Pogonyshev]

Stuart D. Herring wrote:
> > It is probably possible to ask current major mode if we are in a string.
> > If it is not possible, than it is probably worth it to invent a consistent
> > interface that major modes would implement.  I agree that it is not to be
> > done now, of course.  (And while I'm at it, it's a pity that "now" spans
> > the last two years or so.)
> 
> Again, this is not a question of inventiveness, consistency, or any other
> kind of design or programming at all.  It's a mathematical question,
> really: it is easy to ask the question and to devise a mechanism for doing
> so, but it is algorithmically non-trivial to determine the answer.

According to Stefan Monier, you can already ask `syntax-ppss'.  And that
function already does some caching if I was to say from its code.

Paul

Re: transpose-sexps

[Stuart D. Herring]

>> > It is probably possible to ask current major mode if we are in a
>> string.
>> > If it is not possible, than it is probably worth it to invent a
>> consistent
>> > interface that major modes would implement.  I agree that it is not to
>> be
>> > done now, of course.  (And while I'm at it, it's a pity that "now"
>> spans
>> > the last two years or so.)
>>
>> Again, this is not a question of inventiveness, consistency, or any
>> other
>> kind of design or programming at all.  It's a mathematical question,
>> really: it is easy to ask the question and to devise a mechanism for
>> doing
>> so, but it is algorithmically non-trivial to determine the answer.
>
> According to Stefan Monier, you can already ask `syntax-ppss'.  And that
> function already does some caching if I was to say from its code.
>
> Paul

The half of my message you did not include more or less said precisely
that, although I did not name but rather described the function.  The
point is that I don't know whether `syntax-ppss' is always accurate enough
and fast enough for us to want to include it in every call to
`transpose-sexps'; I believe it is still somewhat "beta".

The point is also that the solution you initially proposed ("to invent a
consistent interface that major modes would implement" if there was not
already one) seems to imply that the problem is one of providing a useful
interface to obvious, extant, simple functionality rather than of
implementing/applying a parser with a complicated time-space tradeoff.  I
thought you would want to know that there was a reason for the status quo,
even if it was merely "the alternatives are impractical", instead of it
merely being an oversight or coding bug.

Davis

-- 
This product is sold by volume, not by mass.  If it appears too dense or
too sparse, it is because mass-energy conversion has occurred during
shipping.

Re: transpose-sexps

[Stefan Monnier]

> The half of my message you did not include more or less said precisely
> that, although I did not name but rather described the function.  The
> point is that I don't know whether `syntax-ppss' is always accurate enough
> and fast enough for us to want to include it in every call to
> `transpose-sexps'; I believe it is still somewhat "beta".

It's not even obvious to me that this is what we'd want.  Sometimes, yes,
but I commonly use C-M-t to transpose symbols within strings and comments.


        Stefan