Changes to autoload handling in native-comp branch?

Changes to autoload handling in native-comp branch?

[Alex Bennée]

Hi,

When I did my weekly update of the build I noticed a bunch more errors
being reported when I start up:

  Error loading autoloads: (wrong-type-argument stringp nil) [98 times]

I hesitate to call it breakage as after tweaking a few of my use-package
setups to explicitly load sub-files of a package I'm still able to run
my normal emacs activities. The change was somewhere between 4abb8c822c
and b7678cf10e on the native-comp branch so I was wondering if there are
any pointers for what might have changed so I can dig deeper.

-- 
Alex Bennée

Re: Changes to autoload handling in native-comp branch?

[Andrea Corallo]

Alex Bennée <alex.bennee@linaro.org> writes:

> Hi,
>
> When I did my weekly update of the build I noticed a bunch more errors
> being reported when I start up:
>
>   Error loading autoloads: (wrong-type-argument stringp nil) [98 times]
>
> I hesitate to call it breakage as after tweaking a few of my use-package
> setups to explicitly load sub-files of a package I'm still able to run
> my normal emacs activities. The change was somewhere between 4abb8c822c
> and b7678cf10e on the native-comp branch so I was wondering if there are
> any pointers for what might have changed so I can dig deeper.

Hi Alex,

yeah apparently something broke. I think bug#40620 should be about the
same issue.  I'll have a look.

Thanks!

  Andrea

-- 
akrl@sdf.org

Re: Changes to autoload handling in native-comp branch?

[Andrea Corallo]

Alex Bennée <alex.bennee@linaro.org> writes:

> Hi,
>
> When I did my weekly update of the build I noticed a bunch more errors
> being reported when I start up:
>
>   Error loading autoloads: (wrong-type-argument stringp nil) [98 times]
>
> I hesitate to call it breakage as after tweaking a few of my use-package
> setups to explicitly load sub-files of a package I'm still able to run
> my normal emacs activities. The change was somewhere between 4abb8c822c
> and b7678cf10e on the native-comp branch so I was wondering if there are
> any pointers for what might have changed so I can dig deeper.

Sorry I've missed the second part of your mail.  This weekend I changed
a bunch of things in how we handle file paths for eln files including
the `load-file-name' value but ATM I don't know if and how it is
related.

--
akrl@sdf.org

Re: Changes to autoload handling in native-comp branch?

[Alex Bennée]

Andrea Corallo <akrl@sdf.org> writes:

> Alex Bennée <alex.bennee@linaro.org> writes:
>
>> Hi,
>>
>> When I did my weekly update of the build I noticed a bunch more errors
>> being reported when I start up:
>>
>>   Error loading autoloads: (wrong-type-argument stringp nil) [98 times]
>>
>> I hesitate to call it breakage as after tweaking a few of my use-package
>> setups to explicitly load sub-files of a package I'm still able to run
>> my normal emacs activities. The change was somewhere between 4abb8c822c
>> and b7678cf10e on the native-comp branch so I was wondering if there are
>> any pointers for what might have changed so I can dig deeper.
>
> Hi Alex,
>
> yeah apparently something broke. I think bug#40620 should be about the
> same issue.  I'll have a look.

Yeah manually running that gives me the same error. It all comes down to
the line (file-name-directory #$) which I guess should have some special
handling. The weird thing is on my "normal" emacs (GNU Emacs 25.2.2
(x86_64-pc-linux-gnu) of 2017-09-22, modified by Debian) executing this
line also gives the same error although the *Messages* shows no such
problems during load up.

Does this indicate there us some sort of special context in the way the
autoload code is run? Is file-name-handler-alist modified during
autoload processing in a way that isn't picked up by native code?

>
> Thanks!
>
>   Andrea


-- 
Alex Bennée

Re: Changes to autoload handling in native-comp branch?

[Andrea Corallo]

Alex Bennée <alex.bennee@linaro.org> writes:

> Yeah manually running that gives me the same error. It all comes down to
> the line (file-name-directory #$) which I guess should have some special
> handling. The weird thing is on my "normal" emacs (GNU Emacs 25.2.2
> (x86_64-pc-linux-gnu) of 2017-09-22, modified by Debian) executing this
> line also gives the same error although the *Messages* shows no such
> problems during load up.
>
> Does this indicate there us some sort of special context in the way the
> autoload code is run? Is file-name-handler-alist modified during
> autoload processing in a way that isn't picked up by native code?

Hi Alex,

I confirm was bug#40620 and should be fixed by now.  The bug was
introduced by 6e09597e27.  '$' is replaced by the reader with
`load-file-name' value (and in this branch ATM by load-true-file-name).
This last one I managed to forget to set it were should have been.

Generally speaking I'm not sure having `load-true-file-name' and
`load-file-name' is the nicest design but this was motivated by
bug#40099 (glab :)

https://lists.gnu.org/archive/html/bug-gnu-emacs/2020-04/msg00211.html

Bests

  Andrea

-- 
akrl@sdf.org

Re: Changes to autoload handling in native-comp branch?

[Alex Bennée]

Andrea Corallo <akrl@sdf.org> writes:

> Alex Bennée <alex.bennee@linaro.org> writes:
>
>> Yeah manually running that gives me the same error. It all comes down to
>> the line (file-name-directory #$) which I guess should have some special
>> handling. The weird thing is on my "normal" emacs (GNU Emacs 25.2.2
>> (x86_64-pc-linux-gnu) of 2017-09-22, modified by Debian) executing this
>> line also gives the same error although the *Messages* shows no such
>> problems during load up.
>>
>> Does this indicate there us some sort of special context in the way the
>> autoload code is run? Is file-name-handler-alist modified during
>> autoload processing in a way that isn't picked up by native code?
>
> Hi Alex,
>
> I confirm was bug#40620 and should be fixed by now.  The bug was
> introduced by 6e09597e27.  '$' is replaced by the reader with
> `load-file-name' value (and in this branch ATM by load-true-file-name).
> This last one I managed to forget to set it were should have been.
>
> Generally speaking I'm not sure having `load-true-file-name' and
> `load-file-name' is the nicest design but this was motivated by
> bug#40099 (glab :)
>
> https://lists.gnu.org/archive/html/bug-gnu-emacs/2020-04/msg00211.html

I've updated my tree and can confirm the bug is now gone and things are
working as normal again.

>
> Bests
>
>   Andrea


-- 
Alex Bennée