bug#48079: Temporary files while building after native-comp merge

[Stefan Monnier]

Andrea Corallo [2022-01-07 16:30:05] wrote:
> Stefan Monnier <monnier@iro.umontreal.ca> writes:
>> Andrea Corallo [2022-01-06 16:10:37] wrote:
>>> Eli Zaretskii <eliz@gnu.org> writes:
>>>>> In the plain old byte-compiler, these files are *very* short lived
>>>>> because they're not created during the compilation itself but only at
>>>>> the very end when we save the result to a file (and we just do it by
>>>>> first saving to `foo.elcNNMMPP` and then renaming that to `foo.elc`).
>>>>> 
>>>>> Now with `batch-byte-native-compile-for-bootstrap` apparently we "suspend
>>>>> the byte-compiler" right in the middle of this small time window, i.e. 
>>>>> after
>>>>> writing to `foo.elcNNMMPP` but before its renamed.  Then we call the
>>>>> native compiler and only once the native compiler is done, we resume the
>>>>> byte compilation which just renames the file and exits.
>>>>> 
>>>>> If that understanding is correct, then I think we may be able to fix the
>>>>> problem by just changing the moment at which we suspend the byte-compiler:
>>>>> suspend it *before* it writes to `foo.elcNNMMPP`.
>>>>
>>>> Are you sure your description above is accurate?
>>
>> No, but I haven't heard any indication to the contrary either.
>>
>>>> We have a backtrace in bug#48978 that shows when we create the
>>>> .elcXXX temporary file.  My reading of that backtrace is that it's
>>>> the other way around: native-compilation invokes byte-compile-file,
>>>> which compiles the Lisp into bytecode, creates the file with
>>>> make-temp-file, and writes out the bytecode.  It is true that we then
>>>> defer renaming of the temporary file in this case
>>
>> I don't see why you say "the other way around".  Maybe you're putting
>> more emphasis on some detail of what I wrote than what I intended.
>> The core os what I wrote is that we should change the code so that the
>> native compilation takes place before the `.elcNNMMPP` file is written
>> rather than after.
>>
>> AFAICT the native compiler does not read/need that file because it gets
>> the same information in a more convenient format straight from
>> `bytecomp.el`.
>>
>>> That's correct,
>>
>> I don't know what "That" refers to.
>
> Eli's description of how it works.

But your response seems to say that what I propose can't work, whereas
I can't see in which sense what Eli says contradicts what I say.


        Stefan