[OE-core] subtle weirdness when you combine "_append" with "+="?

[Ulf Magnusson]

On Wed, Aug 17, 2016 at 3:21 PM, Robert P. J. Day <rpjday at crashcourse.ca> wrote:
> On Tue, 16 Aug 2016, Khem Raj wrote:
>
>   ... big snip ...
>
>> the _append/_prepend in conjunction with += is a undocumented
>> behavior however, I do not see it as much a side effect. But in
>> future bitbake may silently change its behavior, so I agree its
>> always good to stay in safe waters.
>
>   so just to finalize this discussion, my thoughts on a reasonable and
> meaningful standard for append/prepend/etc/etc would be to *always*
> use the "mathematical" operators +=/.=/=+/=. except in two situations:
>
>   1) when parsing *must* be delayed until the end, or

I agree as long as that *must* also includes the kind of situation
described for foo.bbclass in
https://www.yoctoproject.org/docs/2.2/bitbake-user-manual/bitbake-user-manual.html#override-style-operation-advantages.
I think you ought to follow the principle of least surprise.

>   2) when combining the operation with an override (for which i'm
>      quite sure there is no equivalent when using +=, etc, right?)
>
> i'm a big fan of simplicity, in the sense that, if an append operation
> does not *require* "_append", it shouldn't use it. because if i see an
> expression:
>
>   VAR_append = "..."
>
> then i immediately wonder if there is a *reason* for using "_append"
> rather than "+=" or ".=", or if that is just the developer's personal
> preference. and i really, really, really hate having to guess.
>
>   oh, and of course, never, ever combine "_append" with "+=" or ".=".
> does that sound like a reasonable coding style?

Sounds reasonable to me. Using _append rather than += can also be
a bug in certain situations. For example, the following does not
guarantee that " value1 value2" appears right after "--flag":

FLAGS = "--flag"
FLAGS_append = " value1"
FLAGS_append = " value2"

Any earlier FLAGS_append value would appear before those two.

Cheers,
Ulf