Re: [bitbake-devel] [PATCH] parse: warn on trailing whitespace after line continuation backslash

On 2026-08-04 13:37, Richard Purdie wrote:
> On Tue, 2026-08-04 at 12:37 +0200, Jaipaul Cheernam wrote:
>>    Hi Richard,
>>
>>     Fair point — it doesn't break parsing itself and the code churn is
>> larger than ideal for a warning.
>>
>>     Would you prefer if I just submitted the oe-core fixes to clean up
>> the affected recipes instead and dropped this bitbake change? That
>> solves the immediate problem without adding complexity to the parser.
> You should definitely send that clean up for oe-core.
   Hi Richard,

   Thanks for the detailed feedback.

I will send a clean-up patch.

> I was toying with the idea we just make whitespace after a line break
> character a fatal parsing error. Most of the fatal errors are inside
> feeder() rather than where you changed the code.
  That would be the cleanest long-term. A fatal error for whitespace 
after \ makes sense since the backslash must be the last character — 
anything after it is always wrong.
> I then looked at the code further and it isn't just trailing whitespace
> after "\" that is an issue. For example, after "}" closing a function
> could also be an issue in the same way. I didn't look in detail beyond
> that, there will be further similar issues though and that does mean
> your patch is incomplete :/.
  Good point, I hadn't considered that.
>>    Alternatively, I could reduce this to a bb.note() (debug-level) so
>> it's only visible with -v, keeping it minimal.
> I think this needs further thought and some wider testing of how
> widespread trailing whitespace is.
>
> We might just want to make any trailing spaces a warning since we
> currently just swallow and hide it unconditionally?
I tried this — warned on any trailing whitespace in all parsed lines 
(comparing raw line minus newline against rstrip'd version). Running 
core-image-minimal:

   - 1665 total warnings (due to repeated re-parse of same files)
   - 59 unique (distinct file:line)
   - 31 files affected (.bbclass, .bb, .inc, .conf)

Most are in bbclass files like sanity.bbclass, buildstats.bbclass, 
useradd.bbclass, kernel-yocto.bbclass, libc-package.bbclass etc. The 
duplication from re-parsing makes it noisy without some form of dedup.

A few options for v2:

  1. Keep it narrow — only warn after \ and } (the cases that actually 
cause patch conflicts). Around 10 unique warnings.
  2. Warn on everything but with dedup (a module-level set). Cleaner 
output but adds state to the parser.
  3. Warn on everything and send a cleanup series fixing all 59 
instances in oe-core. Then the warning stays as a guard for future 
submissions.

What would you prefer?


Thanks,

Jaipaul


>
> Cheers,
>
> Richard
>