`file-attribute' accessors

`file-attribute' accessors

[Lars Magne Ingebrigtsen]

I was reading code that said `(nth 7 (file-attributes file-name))', and
I had to look up the doc string of `file-attributes' and find item seven
to understand what the code was trying to do.

Wouldn't it be nice if we had accessor functions for these things
instead of making people write `nth 7'?  If the code said
`(file-attribute-size (file-attributes file-name))', that would have
made things more understandable.  (And probably easier to write, too,
since one wouldn't have to consult the `file-attributes' doc string
every time...)

If this sounds OK, I can add the, erm, 11 trivial accessor functions and
adjust the documentation of `file-attributes' in trunk.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: `file-attribute' accessors

[Thierry Volpiatto]

Lars Magne Ingebrigtsen <larsi@gnus.org> writes:

> I was reading code that said `(nth 7 (file-attributes file-name))', and
> I had to look up the doc string of `file-attributes' and find item seven
> to understand what the code was trying to do.
>
> Wouldn't it be nice if we had accessor functions for these things
> instead of making people write `nth 7'?  If the code said
> `(file-attribute-size (file-attributes file-name))', that would have
> made things more understandable.  (And probably easier to write, too,
> since one wouldn't have to consult the `file-attributes' doc string
> every time...)
Why not keys instead of many accessor functions?

See `helm-file-attributes' in helm-utils.el at
https://github.com/emacs-helm/helm

(helm-file-attributes "~/.emacs.d/emacs-config-laptop/.emacs.el" :size t)
=>62675
(helm-file-attributes "~/.emacs.d/emacs-config-laptop/.emacs.el" :human-size t)
=>"61.2K"

> If this sounds OK, I can add the, erm, 11 trivial accessor functions and
> adjust the documentation of `file-attributes' in trunk.

-- 
  Thierry
Get my Gnupg key:
gpg --keyserver pgp.mit.edu --recv-keys 59F29997 

Re: `file-attribute' accessors

[Lars Magne Ingebrigtsen]

Thierry Volpiatto <thierry.volpiatto@gmail.com> writes:

> Why not keys instead of many accessor functions?

Because `file-attributes' already has an optional parameter, and one
sometimes wants to have, say, both the uid and the size, and statting
the file twice to get that isn't optimal.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: `file-attribute' accessors

[Thierry Volpiatto]

Lars Magne Ingebrigtsen <larsi@gnus.org> writes:

> Thierry Volpiatto <thierry.volpiatto@gmail.com> writes:
>
>> Why not keys instead of many accessor functions?
>
> Because `file-attributes' already has an optional parameter, and one
> sometimes wants to have, say, both the uid and the size, and statting
> the file twice to get that isn't optimal.
Without key argument i.e only filename, you have all in a proplist, so
you do not have to call function many times to have diverses infos.

-- 
  Thierry
Get my Gnupg key:
gpg --keyserver pgp.mit.edu --recv-keys 59F29997 

Re: `file-attribute' accessors

[Leo]

On 2012-04-09 21:56 +0800, Lars Magne Ingebrigtsen wrote:
> `(file-attribute-size (file-attributes file-name))', that would have

Please name it file-size and possibly file-mtime, file-ctime and
file-atime, which is what I have been doing in my .emacs.

Leo

Re: `file-attribute' accessors

[Ted Zlatanov]

On Mon, 09 Apr 2012 16:40:51 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> Thierry Volpiatto <thierry.volpiatto@gmail.com> writes:
>> Why not keys instead of many accessor functions?

LMI> Because `file-attributes' already has an optional parameter, and one
LMI> sometimes wants to have, say, both the uid and the size, and statting
LMI> the file twice to get that isn't optimal.

I would extend `file-attributes' to take &rest symbols after the extra
format argument, so you could say

(file-attributes FILE nil 'size 'mtime)

And you'd get back a two-element list.  I think that's the most concise
approach: it doesn't create a bunch of new functions, scales from one
attribute to N or "just give me all", and is backwards compatible.

Ted

Re: `file-attribute' accessors

[Lars Magne Ingebrigtsen]

Ted Zlatanov <tzz@lifelogs.com> writes:

> I would extend `file-attributes' to take &rest symbols after the extra
> format argument, so you could say
>
> (file-attributes FILE nil 'size 'mtime)

Yeah, that makes sense.  I kinda like accessor functions, though, since
they make it natural to say stuff like

(reduce '+ (mapcar 'file-attribute-size (mapcar 'file-attributes (directory-files dir))))

but the &rest makes a lot of sense.

While we're bikeshedding, using a plist (as previously suggested) would
also be nice when we're returning lists.  So

(file-attributes FILE nil 'size 'mtime)

could return `(size 3422 mtime (232 42))', perhaps?  Anyway, I think the
optional ID-FORMAT parameter is awkward, and doesn't really belong in
such a low-level access function, in my opinion.  But getting rid of it
is out of the question, I guess?  Although the doc string says

---
ID-FORMAT specifies the preferred format of attributes uid and gid (see
below) - valid values are 'string and 'integer.  The latter is the
default, but we plan to change that, so you should specify a non-nil value
for ID-FORMAT if you use the returned uid or gid.
---

which is going to break stuff, anyway.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: `file-attribute' accessors

[Stefan Monnier]

>> `(file-attribute-size (file-attributes file-name))', that would have
> Please name it file-size and possibly file-mtime, file-ctime and
> file-atime, which is what I have been doing in my .emacs.

Such name would imply that the arg is a file(name) rather than a set of
file-attributes.


        Stefan

Re: `file-attribute' accessors

[Stefan Monnier]

> (file-attributes FILE nil 'size 'mtime)

Too complex.


        Stefan

Re: `file-attribute' accessors

[Stefan Monnier]

> If this sounds OK, I can add the, erm, 11 trivial accessor functions and
> adjust the documentation of `file-attributes' in trunk.

I guess that'd be OK.


        Stefan

Re: `file-attribute' accessors

[Ted Zlatanov]

On Mon, 09 Apr 2012 12:32:09 -0400 Stefan Monnier <monnier@iro.umontreal.ca> wrote: 

>> (file-attributes FILE nil 'size 'mtime)
SM> Too complex.

...as opposed to `(nth 7)' or a separate function for each attribute?
Seriously?

Ted

RE: `file-attribute' accessors

[Drew Adams]

> >> `(file-attribute-size (file-attributes file-name))', that 
> >> would have
> > Please name it file-size and possibly file-mtime, file-ctime and
> > file-atime, which is what I have been doing in my .emacs.
> 
> Such name would imply that the arg is a file(name) rather 
> than a set of file-attributes.

Yes, and `file-attribute-size' gives the impression that it is about the size of
a file attribute.  And so on.

I do not see this requested enhancement (adding 11 attribute accessor functions)
as worth it.  It's misguided - just cruft.

For operations that are particularly useful we could instead add a function that
takes the file name as arg and returns a particular attribute value - e.g.
`file-size', `file-type', `file-uid'.

Other than such cases, YAGNI: If you want only one attribute value there is no
reason to clutter your code with (this-is-the-one (give-me-all-of-them FILE))
instead of just (give-me-the-one FILE).

And we probably do not need such a function for each attribute.  E.g. attribute
#9: "t if file's gid would change if file were deleted and recreated".  YAGNI.

If you prefer, you could append "-attribute" to the name: `file-size-attribute'
etc. instead of `file-size' etc.

But in any case these functions would be applied to a filename, the same way
`file-attributes' is, and not to the result of calling `file-attributes'.
(Accessor functions, indeed - bof.)

> > If this sounds OK, I can add the, erm, 11 trivial accessor 
> > functions and adjust the documentation of `file-attributes'
> > in trunk.
> 
> I guess that'd be OK.

Hasty and bad idea, IMHO.

(If you really want to look for enhancements, check the bug list - there are
lots of enhancement requests there that make more sense than this one.  And
`report-emacs-bug' is the way to submit an enhancement request, in any case.)

Re: `file-attribute' accessors

[Lars Magne Ingebrigtsen]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

>> If this sounds OK, I can add the, erm, 11 trivial accessor functions and
>> adjust the documentation of `file-attributes' in trunk.
>
> I guess that'd be OK.

While writing the doc string to all the functions, it struck me that it
would be kinda nice if the definition of the function would be something
like

(defun file-attribute-size (elem)
  ...
  (if (stringp file)
      (nth 7 (file-attributes elem))
    (nth 7 elem)))

That is, the functions would accept both a string and a file attribute
list, and do the obvious thing in either case.

What do you think?

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: `file-attribute' accessors

[Leo]

On 2012-04-10 00:31 +0800, Stefan Monnier wrote:
> Such name would imply that the arg is a file(name) rather than a set of
> file-attributes.

I didn't realise Lars was proposing accessors on the return value of
file-attributes, which looks kinda cumbersome.

Leo

Re: `file-attribute' accessors

[Stefan Monnier]

> Yes, and `file-attribute-size' gives the impression that it is about
> the size of a file attribute.  And so on.

Yes, that is a bit annoying, although it does correspond to "classic"
Elisp naming.  I personally prefer Scheme style "file-attribute->size".


        Stefan

Re: `file-attribute' accessors

[Stefan Monnier]

>>> (file-attributes FILE nil 'size 'mtime)
SM> Too complex.
> ...as opposed to `(nth 7)' or a separate function for each attribute?
> Seriously?

Yes.


        Stefan

Re: `file-attribute' accessors

[Stefan Monnier]

> (defun file-attribute-size (elem)
>   ...
>   (if (stringp file)
>       (nth 7 (file-attributes elem))
>     (nth 7 elem)))

You mean

  (nth 7 (if (stringp file) ...))

> What do you think?

I'm not sure I like it.  But if we do go this route, then we could name
it `file-size'.


        Stefan

Re: `file-attribute' accessors

[Lars Magne Ingebrigtsen]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

> You mean
>
>   (nth 7 (if (stringp file) ...))

Yes.  :-)

>> What do you think?
>
> I'm not sure I like it.

Yeah, me neither...

> But if we do go this route, then we could name it `file-size'.

For `file-size' in particular, that would make sense, but for
`file-modtime' and `file-uid' (etc) it seems a bit...  odd...  to put
these things that far "up" in the naming hierarchy.

-- 
(domestic pets only, the antidote for overdose, milk.)
  bloggy blog http://lars.ingebrigtsen.no/

Re: `file-attribute' accessors

[Thien-Thi Nguyen]

() Stefan Monnier <monnier@iro.umontreal.ca>
() Mon, 09 Apr 2012 21:47:41 -0400

   Yes, that is a bit annoying, although it does correspond to "classic"
   Elisp naming.  I personally prefer Scheme style "file-attribute->size".

In any case the container is not a single attribute but a set, so to
be excruciatingly correct, the name ought to be ‘file-attributes-size’
(or ‘file-attributes->size’), which is even less pleasant.  :-/

Re: `file-attribute' accessors

[Stefan Monnier]

>    Yes, that is a bit annoying, although it does correspond to "classic"
>    Elisp naming.  I personally prefer Scheme style "file-attribute->size".
> In any case the container is not a single attribute but a set, so to
> be excruciatingly correct, the name ought to be ‘file-attributes-size’
> (or ‘file-attributes->size’),

Yes, sorry, that was a typo.

> which is even less pleasant.  :-/

I don't find it noticeably worse.


        Stefan

Re: `file-attribute' accessors

[Stefan Monnier]

BTW, another approach altogether is to represent file-attributes a bit
differently, e.g. as something like (<type> . <values>), where <type>
(which you could think of as a class if you like OO) describes which
fields exist and at which position, so with a few changes to `get' we
could use (get (file-attributes <file>) 'size).


        Stefan

Re: `file-attribute' accessors

[Miles Bader]

Ted Zlatanov <tzz@lifelogs.com> writes:
> SM> Too complex.
>
> ...as opposed to `(nth 7)' or a separate function for each attribute?
> Seriously?

Trivial accessor functions/macros are _really_ simple, both for humans
and for compilers...

-miles

-- 
古池や 蛙飛び込む 水の音 [松尾芭蕉]