Buffer local variables, let bindings, and read-file-name

Buffer local variables, let bindings, and read-file-name

[Harald Hanche-Olsen]

Is there a good explanation for the following behaviour?

Background: This is on OS X, which has a case insensitive filesystem,
so read-file-name-completion-ignore-case is accordingly set to t.
Possibly, the problem can be reproduced on other OSs by setting the
variable first; but I haven't checked.

----------------
Now if I create a new buffer and evaluate (read-file-name "file: ")
in the buffer, I get case insensitive completion, and all is well.

But if I evaluate (setq-local completion-ignore-case t) in the buffer
and try (read-file-name "file: ") again, completion is case sensitive.
----------------

How can this be? I see that the code for read-file-name-default does
(let ((completion-ignore-case read-file-name-completion-ignore-case)) …)
i.e., it let binds a buffer local variable to the same value it has
already, namely t. And yet, it behaves as if it were now nil.

I know that there can be difficulties with let bindings and
buffer-local variables, but not this kind of difficulties?

Starting emacs with -Q produces the same result.
This is emacs from trunk as of April 11.

- Harald

Re: Buffer local variables, let bindings, and read-file-name

[Michael Heerdegen]

Hello Harald,

> Is there a good explanation for the following behaviour?
>
> Background: This is on OS X, which has a case insensitive filesystem,
> so read-file-name-completion-ignore-case is accordingly set to t.
> Possibly, the problem can be reproduced on other OSs by setting the
> variable first; but I haven't checked.
>
> ----------------
> Now if I create a new buffer and evaluate (read-file-name "file: ")
> in the buffer, I get case insensitive completion, and all is well.
>
> But if I evaluate (setq-local completion-ignore-case t) in the buffer
> and try (read-file-name "file: ") again, completion is case sensitive.
> ----------------
>
> How can this be? I see that the code for read-file-name-default does
> (let ((completion-ignore-case read-file-name-completion-ignore-case)) …)
> i.e., it let binds a buffer local variable to the same value it has
> already, namely t. And yet, it behaves as if it were now nil.
>
> I know that there can be difficulties with let bindings and
> buffer-local variables, but not this kind of difficulties?
>
> Starting emacs with -Q produces the same result.
> This is emacs from trunk as of April 11.

I can reproduce this here with Debian Linux.  I started emacs -Q,
evaluated

  (setq read-file-name-completion-ignore-case t)

(it defaults to nil in my case), and followed your recipe step by step.
And I get exactly the same behavior.  Simply looks like a bug, probably
in the (newly written) completion code.

Please make a bug report (M-x report-emacs-bug), include your recipe and
please also mention that it's the same with Linux.

Sorry for the late reply (where is Stefan?), and thanks for letting us
know.


Regards,

Michael.

Re: Buffer local variables, let bindings, and read-file-name

[Harald Hanche-Olsen]

[Michael Heerdegen <michael_heerdegen@web.de> (2013-05-03 01:14:30 UTC)]

> > ----------------
> > Now if I create a new buffer and evaluate (read-file-name "file: ")
> > in the buffer, I get case insensitive completion, and all is well.
> >
> > But if I evaluate (setq-local completion-ignore-case t) in the buffer
> > and try (read-file-name "file: ") again, completion is case sensitive.
> > ----------------
> 
> I can reproduce this here with Debian Linux.  I started emacs -Q,
> evaluated
> 
>   (setq read-file-name-completion-ignore-case t)
> 
> (it defaults to nil in my case), and followed your recipe step by step.
> And I get exactly the same behavior.  Simply looks like a bug, probably
> in the (newly written) completion code.

Thanks for confirming this.

> Please make a bug report (M-x report-emacs-bug), include your recipe and
> please also mention that it's the same with Linux.

Done. 14340: http://debbugs.gnu.org/cgi/bugreport.cgi?bug=14340

I also added a bit of analysis. It's short enough to repeat here:

read-file-name-default let binds completion-ignore-case to the value of
read-file-name-completion-ignore-case before doing a pile of setup
work and calling completing-read.

When completion-ignore-case is buffer local, let overrides the buffer
local binding and not the global binding. But because completing-read
evaluates the variable in the minibuffer, it gets the global value
instead, i.e., nil.

(I may be wrong about some details in this, but I but this is the gist
of the problem.)

- Harald