On 2018-03-20 14:41, Daniel P. Berrangé wrote:
> On Tue, Mar 20, 2018 at 02:32:16PM +0100, Gerd Hoffmann wrote:
>>   Hi,
>>
>>>> So for these, we should use "".  None of these are generated files though.
>>>
>>> That leads to crazy inconsistent message for developers where 50% of QEMU
>>> header files must use <> and the other 50% of header files must use "".
>>
>> The rules are pretty simple though:
>>
>>    (1) Headers which are generated use <>.
>>    (2) Headers which are in include/ use <>.
>>    (3) Headers sitting in the same directory as the source files use "".
> 
> We have 1200 header files in QEMU - a developer might just reasonably remember
> which header file is a QEMU header, vs which is a 3rd party system header.
> Expecting devs to remember which of those 3 buckets each QEMU header falls
> under is unreasonable IMHO. 

I agree with this in principle, but I don't find it unreasonable for
someone to look up whether a header file exists in the same directory.
I find that much simpler than to look up whether it is a system header,
actually.

And I have to agree with Michael that his rule when to use "" (if the
file is in the same directory) and when to use <> (otherwise) is
actually what every C developer might do by instinct.  (I guess it's
also easier to check in a script than the original guideline?)

However, I also think that if any problem arises because "" was used for
a generated file and that then uses a stale file, the bug is somewhere
else.  And I think that while Michael's proposal is the more intuitive
(C) way, it is a tiny bit more complicated to explain than 'Use "" for
qemu headers, <> for everything else'.

But I guess the main advantage with using this rule I see is that it's
better for people reading the code.  It's just nice to know whether a
file belongs to qemu or not by just looking at the #include statement.
(Note that this implies that it is indeed more difficult to determine
whether a header belongs to qemu than whether it sits in the same
directory as the C file, though!)  So I think the old (current) rule is
better for reading code, Michael's rule would be better for writing
code.  I think reading code is what should be easier.

But since that may be eaten up by build breakages due to stale files, I
don't have a strong opinion either way.  I just wanted to chime in
because in my opinion 'Use "" for headers in the same directory, <> for
everything else' is by no means an unreasonably complicated rule for
people writing code.

Max