[PATCH 0/1] emacs: Pass a copy to notmuch-saved-search-sort-function

[PATCH 0/1] emacs: Pass a copy to notmuch-saved-search-sort-function

[Daniel Schoepe]

As Jani discovered, `sort' modifies its input list, leading to deleted
saved searches when using notmuch-sort-saved-searches.

[PATCH] emacs: Pass a copy to notmuch-saved-search-sort-function

[Daniel Schoepe]

notmuch-saved-search-sort-function might destructively modify its
input (`sort' does that, for instance), so it should not be given
notmuch-saved-searches directly.
---
 emacs/notmuch-hello.el |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/emacs/notmuch-hello.el b/emacs/notmuch-hello.el
index aad373d..b2b22df 100644
--- a/emacs/notmuch-hello.el
+++ b/emacs/notmuch-hello.el
@@ -575,7 +575,10 @@ Complete list of currently available key bindings:
   (let ((searches (notmuch-hello-query-counts
 		   (if notmuch-saved-search-sort-function
 		       (funcall notmuch-saved-search-sort-function
-				notmuch-saved-searches)
+				;; Use a copy, since the sorting
+				;; function may have side effects,
+				;; e.g. if it just `sort's the input.
+				(copy-seq notmuch-saved-searches))
 		     notmuch-saved-searches)
 		   :show-empty-searches notmuch-show-empty-saved-searches))
 	found-target-pos)
-- 
1.7.9.1

[PATCH v2 0/1] emacs: Pass a copy to notmuch-saved-search-sort-function

[Daniel Schoepe]

I overlooked that copy-seq is from cl.el. This version uses copy-sequence instead.

[PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Daniel Schoepe]

notmuch-saved-search-sort-function might destructively modify its
input (`sort' does that, for instance), so it should not be given
notmuch-saved-searches directly.
---
 emacs/notmuch-hello.el |    5 ++++-
 1 files changed, 4 insertions(+), 1 deletions(-)

diff --git a/emacs/notmuch-hello.el b/emacs/notmuch-hello.el
index aad373d..e089290 100644
--- a/emacs/notmuch-hello.el
+++ b/emacs/notmuch-hello.el
@@ -575,7 +575,10 @@ Complete list of currently available key bindings:
   (let ((searches (notmuch-hello-query-counts
 		   (if notmuch-saved-search-sort-function
 		       (funcall notmuch-saved-search-sort-function
-				notmuch-saved-searches)
+				;; Use a copy, since the sorting
+				;; function may have side effects,
+				;; e.g. if it just `sort's the input.
+				(copy-sequence notmuch-saved-searches))
 		     notmuch-saved-searches)
 		   :show-empty-searches notmuch-show-empty-saved-searches))
 	found-target-pos)
-- 
1.7.9.1

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Jani Nikula]

On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <daniel@schoepe.org> wrote:
> notmuch-saved-search-sort-function might destructively modify its
> input (`sort' does that, for instance), so it should not be given
> notmuch-saved-searches directly.

Hi Daniel, thanks for fixing this. Works for me.

Full disclosure: I put the broken sort usage there in the first place...

BR,
Jani.


> ---
>  emacs/notmuch-hello.el |    5 ++++-
>  1 files changed, 4 insertions(+), 1 deletions(-)
> 
> diff --git a/emacs/notmuch-hello.el b/emacs/notmuch-hello.el
> index aad373d..e089290 100644
> --- a/emacs/notmuch-hello.el
> +++ b/emacs/notmuch-hello.el
> @@ -575,7 +575,10 @@ Complete list of currently available key bindings:
>    (let ((searches (notmuch-hello-query-counts
>  		   (if notmuch-saved-search-sort-function
>  		       (funcall notmuch-saved-search-sort-function
> -				notmuch-saved-searches)
> +				;; Use a copy, since the sorting
> +				;; function may have side effects,
> +				;; e.g. if it just `sort's the input.
> +				(copy-sequence notmuch-saved-searches))
>  		     notmuch-saved-searches)
>  		   :show-empty-searches notmuch-show-empty-saved-searches))
>  	found-target-pos)
> -- 
> 1.7.9.1
> 
> _______________________________________________
> notmuch mailing list
> notmuch@notmuchmail.org
> http://notmuchmail.org/mailman/listinfo/notmuch

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Dmitry Kurochkin]

On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <daniel@schoepe.org> wrote:
> notmuch-saved-search-sort-function might destructively modify its
> input (`sort' does that, for instance), so it should not be given
> notmuch-saved-searches directly.
> ---

-1

I think we should require `notmuch-saved-search-sort-function' not to
have side effects.  Current documentation should be more clear about
this.  We need to fix `notmuch-sort-saved-searches' to copy the list
before calling `sort'.  But we should not do it in
`notmuch-hello-insert-saved-searches' for any sorting function (which
may not need this copying).

Regards,
  Dmitry

>  emacs/notmuch-hello.el |    5 ++++-
>  1 files changed, 4 insertions(+), 1 deletions(-)
> 
> diff --git a/emacs/notmuch-hello.el b/emacs/notmuch-hello.el
> index aad373d..e089290 100644
> --- a/emacs/notmuch-hello.el
> +++ b/emacs/notmuch-hello.el
> @@ -575,7 +575,10 @@ Complete list of currently available key bindings:
>    (let ((searches (notmuch-hello-query-counts
>  		   (if notmuch-saved-search-sort-function
>  		       (funcall notmuch-saved-search-sort-function
> -				notmuch-saved-searches)
> +				;; Use a copy, since the sorting
> +				;; function may have side effects,
> +				;; e.g. if it just `sort's the input.
> +				(copy-sequence notmuch-saved-searches))
>  		     notmuch-saved-searches)
>  		   :show-empty-searches notmuch-show-empty-saved-searches))
>  	found-target-pos)
> -- 
> 1.7.9.1
> 
> _______________________________________________
> notmuch mailing list
> notmuch@notmuchmail.org
> http://notmuchmail.org/mailman/listinfo/notmuch

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Daniel Schoepe]

[-- Attachment #1: Type: text/plain, Size: 1459 bytes --]

On Mon, 05 Mar 2012 06:21:52 +0400, Dmitry Kurochkin <dmitry.kurochkin@gmail.com> wrote:
> On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <daniel@schoepe.org> wrote:
> > notmuch-saved-search-sort-function might destructively modify its
> > input (`sort' does that, for instance), so it should not be given
> > notmuch-saved-searches directly.
> > ---
> 
> -1
> 
> I think we should require `notmuch-saved-search-sort-function' not to
> have side effects.  Current documentation should be more clear about
> this.  We need to fix `notmuch-sort-saved-searches' to copy the list
> before calling `sort'.  But we should not do it in
> `notmuch-hello-insert-saved-searches' for any sorting function (which
> may not need this copying).

My reasoning was that since sort is such a common function, many users
will probably use sort for their own sorting functions, not realizing
that it has side effects. This will lead to confusing behavior that's
not so easy to track down.

Copying the list of saved searches when running notmuch-hello does not
seem be relevant to performance to me, since it's a) not called that
often and b) the list of saved searches will rarely exceed 30 elements.

Hence, this way we can avoid some headaches for users who define their
own sorting functions at a negligible (performance) cost. Incidentally,
this is also how notmuch-hello did it before the user-defined sections
patches.

Cheers,
Daniel

[-- Attachment #2: Type: application/pgp-signature, Size: 835 bytes --]

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Tomi Ollila]

On Mon, 05 Mar 2012 12:17:43 +0100, Daniel Schoepe <daniel@schoepe.org> wrote:
> On Mon, 05 Mar 2012 06:21:52 +0400, Dmitry Kurochkin <dmitry.kurochkin@gmail.com> wrote:
> > On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <daniel@schoepe.org> wrote:
> > > notmuch-saved-search-sort-function might destructively modify its
> > > input (`sort' does that, for instance), so it should not be given
> > > notmuch-saved-searches directly.
> > > ---
> > 
> > -1
> > 
> > I think we should require `notmuch-saved-search-sort-function' not to
> > have side effects.  Current documentation should be more clear about
> > this.  We need to fix `notmuch-sort-saved-searches' to copy the list
> > before calling `sort'.  But we should not do it in
> > `notmuch-hello-insert-saved-searches' for any sorting function (which
> > may not need this copying).
> 
> My reasoning was that since sort is such a common function, many users
> will probably use sort for their own sorting functions, not realizing
> that it has side effects. This will lead to confusing behavior that's
> not so easy to track down.
> 
> Copying the list of saved searches when running notmuch-hello does not
> seem be relevant to performance to me, since it's a) not called that
> often and b) the list of saved searches will rarely exceed 30 elements.
> 
> Hence, this way we can avoid some headaches for users who define their
> own sorting functions at a negligible (performance) cost. Incidentally,
> this is also how notmuch-hello did it before the user-defined sections
> patches.

Hard to say -- maybe the alternative:

 (defun notmuch-sort-saved-searches (alist)
   "Generate an alphabetically sorted saved searches alist."
-  (sort alist (lambda (a b) (string< (car a) (car b)))))
+  (sort (copy-sequence alist) (lambda (a b) (string< (car a) (car b)))))

matches better with the current documentation 
(of notmuch-saved-search-sort-function).

Both sort and copy-sequence are blazingly fast...

For more information, read
http://www.emacswiki.org/emacs/DestructiveOperations

> Cheers,
> Daniel

Tomi

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Dmitry Kurochkin]

On Mon, 05 Mar 2012 12:17:43 +0100, Daniel Schoepe <daniel@schoepe.org> wrote:
> On Mon, 05 Mar 2012 06:21:52 +0400, Dmitry Kurochkin <dmitry.kurochkin@gmail.com> wrote:
> > On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <daniel@schoepe.org> wrote:
> > > notmuch-saved-search-sort-function might destructively modify its
> > > input (`sort' does that, for instance), so it should not be given
> > > notmuch-saved-searches directly.
> > > ---
> > 
> > -1
> > 
> > I think we should require `notmuch-saved-search-sort-function' not to
> > have side effects.  Current documentation should be more clear about
> > this.  We need to fix `notmuch-sort-saved-searches' to copy the list
> > before calling `sort'.  But we should not do it in
> > `notmuch-hello-insert-saved-searches' for any sorting function (which
> > may not need this copying).
> 
> My reasoning was that since sort is such a common function, many users
> will probably use sort for their own sorting functions, not realizing
> that it has side effects. This will lead to confusing behavior that's
> not so easy to track down.
> 
> Copying the list of saved searches when running notmuch-hello does not
> seem be relevant to performance to me, since it's a) not called that
> often and b) the list of saved searches will rarely exceed 30 elements.
> 
> Hence, this way we can avoid some headaches for users who define their
> own sorting functions at a negligible (performance) cost. Incidentally,
> this is also how notmuch-hello did it before the user-defined sections
> patches.
> 

I do not buy the argument that we should help users who implement their
own sorting functions but do not read documentation for functions they
use.  Apparently, those who implemented the `sort' function had similar
ideas.  And I do not think it is our job to add workarounds for it.

An alternative (and IMO better) solution would be to allow customization
of compare function used for sorting instead of the sorting function
itself.

Regards,
  Dmitry

> Cheers,
> Daniel

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Jani Nikula]

[-- Attachment #1: Type: text/plain, Size: 2847 bytes --]

On Mar 5, 2012 5:43 PM, "Dmitry Kurochkin" <dmitry.kurochkin@gmail.com>
wrote:
>
> On Mon, 05 Mar 2012 12:17:43 +0100, Daniel Schoepe <daniel@schoepe.org>
wrote:
> > On Mon, 05 Mar 2012 06:21:52 +0400, Dmitry Kurochkin <
dmitry.kurochkin@gmail.com> wrote:
> > > On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <daniel@schoepe.org>
wrote:
> > > > notmuch-saved-search-sort-function might destructively modify its
> > > > input (`sort' does that, for instance), so it should not be given
> > > > notmuch-saved-searches directly.
> > > > ---
> > >
> > > -1
> > >
> > > I think we should require `notmuch-saved-search-sort-function' not to
> > > have side effects.  Current documentation should be more clear about
> > > this.  We need to fix `notmuch-sort-saved-searches' to copy the list
> > > before calling `sort'.  But we should not do it in
> > > `notmuch-hello-insert-saved-searches' for any sorting function (which
> > > may not need this copying).
> >
> > My reasoning was that since sort is such a common function, many users
> > will probably use sort for their own sorting functions, not realizing
> > that it has side effects. This will lead to confusing behavior that's
> > not so easy to track down.
> >
> > Copying the list of saved searches when running notmuch-hello does not
> > seem be relevant to performance to me, since it's a) not called that
> > often and b) the list of saved searches will rarely exceed 30 elements.
> >
> > Hence, this way we can avoid some headaches for users who define their
> > own sorting functions at a negligible (performance) cost. Incidentally,
> > this is also how notmuch-hello did it before the user-defined sections
> > patches.
> >
>
> I do not buy the argument that we should help users who implement their
> own sorting functions but do not read documentation for functions they
> use.  Apparently, those who implemented the `sort' function had similar
> ideas.  And I do not think it is our job to add workarounds for it.
>
> An alternative (and IMO better) solution would be to allow customization
> of compare function used for sorting instead of the sorting function
> itself.

Providing the customization of the sort function is more powerful than the
compare function. In the case of saved searches I can imagine people might
want to partially use the original order while sort the rest (e.g.
important ones first in predefined order, others sorted). In fact this also
allows dropping out some elements. And renaming. And changing the queries...

(I had something like that in mind originally but then settled with just
capitalizing the important ones to show them first.)

BR,
Jani.

>
> Regards,
>  Dmitry
>
> > Cheers,
> > Daniel
> _______________________________________________
> notmuch mailing list
> notmuch@notmuchmail.org
> http://notmuchmail.org/mailman/listinfo/notmuch

[-- Attachment #2: Type: text/html, Size: 3810 bytes --]

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Daniel Schoepe]

[-- Attachment #1: Type: text/plain, Size: 1220 bytes --]

On Mar 5, 2012 5:43 PM, "Dmitry Kurochkin" <dmitry.kurochkin@gmail.com>
> I do not buy the argument that we should help users who implement their
> own sorting functions but do not read documentation for functions they
> use.  Apparently, those who implemented the `sort' function had similar
> ideas.  And I do not think it is our job to add workarounds for it.

Yeah that also sounds a bit more consistent, I guess my perception was
colored by my surprise at learning that sort has side effects in elisp. :)

On Mon, 5 Mar 2012 22:55:54 +0200, Jani Nikula <jani@nikula.org> wrote:
> Providing the customization of the sort function is more powerful than the
> compare function. In the case of saved searches I can imagine people might
> want to partially use the original order while sort the rest (e.g.
> important ones first in predefined order, others sorted). In fact this also
> allows dropping out some elements. And renaming. And changing the queries...
> 
> (I had something like that in mind originally but then settled with just
> capitalizing the important ones to show them first.)

I have the same objections to only providing a customization option for
comparing.

Cheers,
Daniel

[-- Attachment #2: Type: application/pgp-signature, Size: 835 bytes --]

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Dmitry Kurochkin]

On Mon, 5 Mar 2012 22:55:54 +0200, Jani Nikula <jani@nikula.org> wrote:
> On Mar 5, 2012 5:43 PM, "Dmitry Kurochkin" <dmitry.kurochkin@gmail.com>
> wrote:
> >
> > On Mon, 05 Mar 2012 12:17:43 +0100, Daniel Schoepe <daniel@schoepe.org>
> wrote:
> > > On Mon, 05 Mar 2012 06:21:52 +0400, Dmitry Kurochkin <
> dmitry.kurochkin@gmail.com> wrote:
> > > > On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <daniel@schoepe.org>
> wrote:
> > > > > notmuch-saved-search-sort-function might destructively modify its
> > > > > input (`sort' does that, for instance), so it should not be given
> > > > > notmuch-saved-searches directly.
> > > > > ---
> > > >
> > > > -1
> > > >
> > > > I think we should require `notmuch-saved-search-sort-function' not to
> > > > have side effects.  Current documentation should be more clear about
> > > > this.  We need to fix `notmuch-sort-saved-searches' to copy the list
> > > > before calling `sort'.  But we should not do it in
> > > > `notmuch-hello-insert-saved-searches' for any sorting function (which
> > > > may not need this copying).
> > >
> > > My reasoning was that since sort is such a common function, many users
> > > will probably use sort for their own sorting functions, not realizing
> > > that it has side effects. This will lead to confusing behavior that's
> > > not so easy to track down.
> > >
> > > Copying the list of saved searches when running notmuch-hello does not
> > > seem be relevant to performance to me, since it's a) not called that
> > > often and b) the list of saved searches will rarely exceed 30 elements.
> > >
> > > Hence, this way we can avoid some headaches for users who define their
> > > own sorting functions at a negligible (performance) cost. Incidentally,
> > > this is also how notmuch-hello did it before the user-defined sections
> > > patches.
> > >
> >
> > I do not buy the argument that we should help users who implement their
> > own sorting functions but do not read documentation for functions they
> > use.  Apparently, those who implemented the `sort' function had similar
> > ideas.  And I do not think it is our job to add workarounds for it.
> >
> > An alternative (and IMO better) solution would be to allow customization
> > of compare function used for sorting instead of the sorting function
> > itself.
> 
> Providing the customization of the sort function is more powerful than the
> compare function. In the case of saved searches I can imagine people might
> want to partially use the original order while sort the rest (e.g.
> important ones first in predefined order, others sorted).

Valid point.

> In fact this also
> allows dropping out some elements. And renaming. And changing the queries...
> 
> (I had something like that in mind originally but then settled with just
> capitalizing the important ones to show them first.)
> 

All of these are invalid usages of `notmuch-saved-search-sort-function'.
The function is meant for sorting only (hence the name).  So the code
might assume that the function does only sorting.

I do not understand why we need such functionality (renaming,
capitalizing, etc.).  You can just rename the query itself if you want
to.  Should be easier IMO.  But if we need such functionality, we should
not misuse sorting function for it.  We can add `notmuch-saved-searches'
function which would return saved searches list (sorted, renamed and
mangled in any other way).  By default it would return
`notmuch-saved-searches' variable as is.

Regards,
  Dmitry

> BR,
> Jani.
> 
> >
> > Regards,
> >  Dmitry
> >
> > > Cheers,
> > > Daniel
> > _______________________________________________
> > notmuch mailing list
> > notmuch@notmuchmail.org
> > http://notmuchmail.org/mailman/listinfo/notmuch
Non-text part: text/html

Re: [PATCH v2] emacs: Pass a copy to notmuch-saved-search-sort-function

[Jani Nikula]

[-- Attachment #1: Type: text/plain, Size: 4066 bytes --]

On Mar 5, 2012 11:11 PM, "Dmitry Kurochkin" <dmitry.kurochkin@gmail.com>
wrote:
>
> On Mon, 5 Mar 2012 22:55:54 +0200, Jani Nikula <jani@nikula.org> wrote:
> > On Mar 5, 2012 5:43 PM, "Dmitry Kurochkin" <dmitry.kurochkin@gmail.com>
> > wrote:
> > >
> > > On Mon, 05 Mar 2012 12:17:43 +0100, Daniel Schoepe <daniel@schoepe.org
>
> > wrote:
> > > > On Mon, 05 Mar 2012 06:21:52 +0400, Dmitry Kurochkin <
> > dmitry.kurochkin@gmail.com> wrote:
> > > > > On Thu,  1 Mar 2012 21:24:38 +0100, Daniel Schoepe <
daniel@schoepe.org>
> > wrote:
> > > > > > notmuch-saved-search-sort-function might destructively modify
its
> > > > > > input (`sort' does that, for instance), so it should not be
given
> > > > > > notmuch-saved-searches directly.
> > > > > > ---
> > > > >
> > > > > -1
> > > > >
> > > > > I think we should require `notmuch-saved-search-sort-function'
not to
> > > > > have side effects.  Current documentation should be more clear
about
> > > > > this.  We need to fix `notmuch-sort-saved-searches' to copy the
list
> > > > > before calling `sort'.  But we should not do it in
> > > > > `notmuch-hello-insert-saved-searches' for any sorting function
(which
> > > > > may not need this copying).
> > > >
> > > > My reasoning was that since sort is such a common function, many
users
> > > > will probably use sort for their own sorting functions, not
realizing
> > > > that it has side effects. This will lead to confusing behavior
that's
> > > > not so easy to track down.
> > > >
> > > > Copying the list of saved searches when running notmuch-hello does
not
> > > > seem be relevant to performance to me, since it's a) not called that
> > > > often and b) the list of saved searches will rarely exceed 30
elements.
> > > >
> > > > Hence, this way we can avoid some headaches for users who define
their
> > > > own sorting functions at a negligible (performance) cost.
Incidentally,
> > > > this is also how notmuch-hello did it before the user-defined
sections
> > > > patches.
> > > >
> > >
> > > I do not buy the argument that we should help users who implement
their
> > > own sorting functions but do not read documentation for functions they
> > > use.  Apparently, those who implemented the `sort' function had
similar
> > > ideas.  And I do not think it is our job to add workarounds for it.
> > >
> > > An alternative (and IMO better) solution would be to allow
customization
> > > of compare function used for sorting instead of the sorting function
> > > itself.
> >
> > Providing the customization of the sort function is more powerful than
the
> > compare function. In the case of saved searches I can imagine people
might
> > want to partially use the original order while sort the rest (e.g.
> > important ones first in predefined order, others sorted).
>
> Valid point.
>
> > In fact this also
> > allows dropping out some elements. And renaming. And changing the
queries...
> >
> > (I had something like that in mind originally but then settled with just
> > capitalizing the important ones to show them first.)
> >
>
> All of these are invalid usages of `notmuch-saved-search-sort-function'.
> The function is meant for sorting only (hence the name).  So the code
> might assume that the function does only sorting.
>
> I do not understand why we need such functionality (renaming,
> capitalizing, etc.).  You can just rename the query itself if you want
> to.  Should be easier IMO.

Just for the record, I have a few important searches capitalized in the
saved searches and just use the regular sort. Capitalized entries sort
before the lowercase ones.

> But if we need such functionality, we should
> not misuse sorting function for it.  We can add `notmuch-saved-searches'
> function which would return saved searches list (sorted, renamed and
> mangled in any other way).  By default it would return
> `notmuch-saved-searches' variable as is.

Agreed.

As to the problem at hand, we should just fix the sort function not to
modify its input. I wouldn't hold my breath waiting for someone to provide
patches for the rest...

BR,
Jani.

[-- Attachment #2: Type: text/html, Size: 5536 bytes --]