Byte-compilation warning in eldoc.el

Byte-compilation warning in eldoc.el

[Eli Zaretskii]

  In toplevel form:
  emacs-lisp/eldoc.el:540:2: Warning: unintern called with 1 argument, but
      requires 2

Re: Byte-compilation warning in eldoc.el

[João Távora]

Eli Zaretskii <eliz@gnu.org> writes:

>   In toplevel form:
>   emacs-lisp/eldoc.el:540:2: Warning: unintern called with 1 argument, but
>       requires 2

Thanks, fixed.  But curiously it doesn't.  You can call it (and I always
do call it) with one argument, as one does in CL.

By the way, the only reason I had to do this (I rarely use this outside
interactive contexts) is to make eldoc.el work in Emacs 26.3.  I explain
this in the commit message, but I'm still uncertain of the cause of the
problem:

Author: João Távora <joaotavora@gmail.com>
Date:   Fri Jul 10 01:47:10 2020 +0100

    Fix Eldoc problem when loading on Emacs 26.3
    
    When defining the obsolete variable alias for old
    eldoc-documentation-function (which now points to the newer
    eldoc-documentation-strategy), one gets the error "don't know how to
    make a localized vareiable an alias".  I'm not sure, but I suspect
    this is because Eldoc is preloaded in Emacs 26.3 and the
    eldoc-documentation-function variable is already set locally by some
    Elisp buffer.
    
    Uninterning the symbol shortly before defining the alias seems to fix
    it.

Separately, I also noticed that calling `package-install` for Eldoc in
Emacs 26.3, will _not_ load eldoc, and neither will `require`.  One has
to `load` it specifically, even after calling `package-initialize`.

João

Re: Byte-compilation warning in eldoc.el

[Eli Zaretskii]

> From: João Távora <joaotavora@gmail.com>
> Cc: emacs-devel@gnu.org
> Date: Fri, 10 Jul 2020 08:20:37 +0100
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> >   In toplevel form:
> >   emacs-lisp/eldoc.el:540:2: Warning: unintern called with 1 argument, but
> >       requires 2
> 
> Thanks, fixed.  But curiously it doesn't.  You can call it (and I always
> do call it) with one argument, as one does in CL.

That's because we want the optional 2nd argument be explicitly
mentioned.

Re: Byte-compilation warning in eldoc.el

[Noam Postavsky]

On Fri, 10 Jul 2020 at 03:20, João Távora <joaotavora@gmail.com> wrote:
>     When defining the obsolete variable alias for old
>     eldoc-documentation-function (which now points to the newer
>     eldoc-documentation-strategy), one gets the error "don't know how to
>     make a localized vareiable an alias".  I'm not sure, but I suspect
>     this is because Eldoc is preloaded in Emacs 26.3 and the
>     eldoc-documentation-function variable is already set locally by some
>     Elisp buffer.
>
>     Uninterning the symbol shortly before defining the alias seems to fix
>     it.

makunbound might fit better.

Re: Byte-compilation warning in eldoc.el

[Stefan Monnier]

> Thanks, fixed.  But curiously it doesn't.  You can call it (and I always
> do call it) with one argument, as one does in CL.

Technically, you can call it with one argument.
But in 99% of the cases it's a really bad idea.

> By the way, the only reason I had to do this (I rarely use this outside
> interactive contexts)

Exactly: it might make sense occasionally while debugging, but is
usually an error in code.
[ Using it with a second argument is quite different, because then it
  usually corresponds to a simple `remhash`.  ]

>     When defining the obsolete variable alias for old
>     eldoc-documentation-function (which now points to the newer
>     eldoc-documentation-strategy),

Which reminds me: I think it'd be better not to make it an alias.

> Separately, I also noticed that calling `package-install` for Eldoc in
> Emacs 26.3, will _not_ load eldoc, and neither will `require`.

Not sure what you mean with the "_not_ load" part, but as for `require`,
it's normal because it's preloaded.


        Stefan

RE: Byte-compilation warning in eldoc.el

[Drew Adams]

> > Thanks, fixed.  But curiously it doesn't.  You can call it (and I
> > always do call it) with one argument, as one does in CL.
> 
> Technically, you can call it with one argument.
> But in 99% of the cases it's a really bad idea.

Why?  What are those 99% (or what is the 1%)?
What's the (undocumented) behavior of calling it
with no second arg?  Isn't it just to use the
value of `obarray' (the standard obarray)?

What's the reason that the 2nd arg isn't called
out as optional, with the doc string saying that
it defaults to the value of variable `obarray'?

What's good about having an optional parameter
that's not advertised as such?  What's the real
point here, that calls for this exception?

Re: Byte-compilation warning in eldoc.el

[João Távora]

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

>> Thanks, fixed.  But curiously it doesn't.  You can call it (and I always
>> do call it) with one argument, as one does in CL.
>
> Technically, you can call it with one argument.
> But in 99% of the cases it's a really bad idea.

What's the difference to `intern'? or `intern-soft'?

>> By the way, the only reason I had to do this (I rarely use this outside
>> interactive contexts)
>
> Exactly: it might make sense occasionally while debugging, but is
> usually an error in code.

Not arguing but, like Drew, I wonder why.  FWIW in Common Lisp, intern,
unintern, find-symbol, etc all assume the "current package" as the thing
to operate on.

>> Separately, I also noticed that calling `package-install` for Eldoc in
>> Emacs 26.3, will _not_ load eldoc, and neither will `require`.
>
> Not sure what you mean with the "_not_ load" part, but as for `require`,
> it's normal because it's preloaded.

Sorry, it means two different things:

1. when you `package-install` the new eldoc 1.2.0 package in Emacs 26.3
   for the first time, it just compiles it, it doesn't load it.  We had
   witnessed this with Flymake too, remember?

2. When you shut down that Emacs session, start it again, run
   package-initialize, you _still_ don't have the new eldoc.el file
   (1.2.0) loaded.  This is only for eldoc.el, flymake.el works
   correctly in this use case.

João

Re: Byte-compilation warning in eldoc.el

[Stefan Monnier]

>>> Thanks, fixed.  But curiously it doesn't.  You can call it (and I always
>>> do call it) with one argument, as one does in CL.
>> Technically, you can call it with one argument.
>> But in 99% of the cases it's a really bad idea.
> What's the difference to `intern'? or `intern-soft'?

Interning a new symbol doesn't affect existing ones, and interning an
old symbol doesn't do anything, so those two are perfectly harmless.
`unintern` on the other hand can render your Emacs session inoperable.

>>> By the way, the only reason I had to do this (I rarely use this outside
>>> interactive contexts)
>> Exactly: it might make sense occasionally while debugging, but is
>> usually an error in code.
> Not arguing but, like Drew, I wonder why.  FWIW in Common Lisp, intern,
> unintern, find-symbol, etc all assume the "current package" as the thing
> to operate on.

Same here.  `unintern` on the global obarray is really dangerous and
very rarely needed.  Most cases I've found were misuses better served
by something different (e.g. `makunbound`).

> 1. when you `package-install` the new eldoc 1.2.0 package in Emacs 26.3
>    for the first time, it just compiles it, it doesn't load it.  We had
>    witnessed this with Flymake too, remember?

`package-install` is not supposed to "load" the package (it installs
it, activates it, and compiles it).  But yes, it is also supposed to
*reload* already loaded files (right before compiling), so it looks like
there might be a bug somewhere.

You might want to trace through `package--load-files-for-activation` to
see why eldoc.el is not reloaded even though it should be in
`load-history`.

> 2. When you shut down that Emacs session, start it again, run
>    package-initialize, you _still_ don't have the new eldoc.el file
>    (1.2.0) loaded.  This is only for eldoc.el, flymake.el works
>    correctly in this use case.

Yes, this is because it's preloaded in Emacs, so when package.el comes
to activate the package, the autoloads in `eldoc-autoloads.el` won't have any
effect because all its functions are already defined (since `eldoc.el`
is already loaded).

This is not a really bug in package.el but arguably a missing feature,
indeed.  When `package-activate-all` is called, it should probably
reload all packages that were already loaded before (when the version
already loaded is different from the one that's just been activated).


        Stefan

Re: Byte-compilation warning in eldoc.el

[João Távora]

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

>> 1. when you `package-install` the new eldoc 1.2.0 package in Emacs 26.3
>>    for the first time, it just compiles it, it doesn't load it.  We had
>>    witnessed this with Flymake too, remember?
>
> `package-install` is not supposed to "load" the package (it installs
> it, activates it, and compiles it).  But yes, it is also supposed to
> *reload* already loaded files (right before compiling), so it looks like
> there might be a bug somewhere.
>
> You might want to trace through `package--load-files-for-activation` to
> see why eldoc.el is not reloaded even though it should be in
> `load-history`.
>
>> 2. When you shut down that Emacs session, start it again, run
>>    package-initialize, you _still_ don't have the new eldoc.el file
>>    (1.2.0) loaded.  This is only for eldoc.el, flymake.el works
>>    correctly in this use case.
>
> Yes, this is because it's preloaded in Emacs, so when package.el comes
> to activate the package, the autoloads in `eldoc-autoloads.el` won't have any
> effect because all its functions are already defined (since `eldoc.el`
> is already loaded).
>
> This is not a really bug in package.el but arguably a missing feature,
> indeed.  When `package-activate-all` is called, it should probably
> reload all packages that were already loaded before (when the version
> already loaded is different from the one that's just been activated).

Sounds reasoable.  But can these changes ever reach Emacs 26.3?  How?
Only by updating package.el itself, right?  Is that possible?

I'm afraid I don't have time to do either of those things, right now, so
I'm just going to hack it in Eglot, for now.  But I'm also reporting
this as a bug, so we can track it.

João

Re: Byte-compilation warning in eldoc.el

[Stefan Monnier]

>> This is not a really bug in package.el but arguably a missing feature,
>> indeed.  When `package-activate-all` is called, it should probably
>> reload all packages that were already loaded before (when the version
>> already loaded is different from the one that's just been activated).
> Sounds reasoable.  But can these changes ever reach Emacs 26.3?

Probably not, no, but in a few years nobody will care.
The important thing is to fix it in `master` so that in a few years the
problem is solved.


        Stefan