[PATCH 1/3] waf_generator: Copy headers if necessary.

[Chris Johns]

On 10/08/2017 16:06, Christian Mauderer wrote:
> Am 10.08.2017 um 07:55 schrieb Chris Johns:
>> On 10/08/2017 15:34, Christian Mauderer wrote:
>>> Am 10.08.2017 um 01:08 schrieb Chris Johns:
>>>> On 10/08/2017 02:29, Sichen Zhao wrote:
>>>>> From: Christian Mauderer <christian.mauderer at embedded-brains.de>
>>> There was no real reason. I just chose an arbitrary name.
>>> 'build-include' is fine for me too. I would avoid just 'include' because
>>> it's a very generic name. If it is added somewhere else, that might lead
>>> to a conflict.
>>
>> Ah yes, 'build-include' is fine.
>>
>>>>> +        start_dir = bld.path.find_dir(headers[0])
>>>>> +        for header in start_dir.ant_glob("**/" + headers[1]):
>>>>
>>>> Do we always want to copy all the files?
>>>>
>>>
>>> What do you mean by "all the files"? As far as I understood, only the
>>> files that are changed should be copied by waf.
>>>
>>> Or do you mean the ant_glob? In that case: I modeled that after the
>>> install target of the headers. 
>>
>> Should this be an ant_glob or a list? I am wondering about needing to copy a
>> selected number of files from a group a glob would catch. If this is not
>> happening that is fine, it can be added if the situation arises.
>>
> 
> Oh, I think I start to understand what you mean. Like already said, this
> works around the problem that the headers are not installed when the
> sources are built. So it should definitely should copy the same files
> like the install part that also uses that array.

Agreed.

> It is possible to add a glob which catches only a single file (even
> better if we move the "**" into builder.py). You might have noted that
> Sichen had to add a lot of such lines for openssl. It would be nice to
> have some cleaner method of writing that but I haven't found one yet.

Yes I had noticed it is a large package. Is the number of files in the libbsd.py
module the concern? There is a point of view that says explicit listing of files
is better because you do not accidentally pick up files you should not where
glob'ing a tree in builder.py or libbsd.py could do that.

The include lists in builder.py was something I knew would come back and bite
me. There was only so much I could do at the time and this fell of the end. I
thought the list might grow but could not see a real use case. I can fix this
once the patch is merged.

> Maybe you have an idea?

Not at the moment. Lets get this work in and then I can take a look. It will be
simpler and also clearer to see what is happening.

I could look at making libbsd.py into libbsd/__init__.py with a file per module
and the top level just sees the imports. I still see no need for us to have
module build control, we always build all modules.

>>>> If a new field is not added can you please update builder.py with this rule so
>>>> we know what to do when adding headers to builder.headerPaths()?
>>>
>>> You mean some comment that describes that behavior? I can add that also
>>> I don't really think that it is necessary. It's just a workaround so
>>> that we can build the binaries without having to install the headers
>>> before the build.
>>
>> It took me a while to see what was happening. The table in builder.py is a long
>> way from here so even a note would help.
> 
> OK. I'll add a comment that shortly describes what happens with the list
> of header files.
> 

Thanks.

Chris