ohos: Use custom domain when logging with hilog

[jschwe]

This modifies the existing patch to redirect log messages to hilog to use a non-zero domain, which allows us to filter logs via the domain on ohos devices.
We set 0xE0C3 for Rust code in servo, using a different number allows us to seperate spidermonkey from servo logs if wanted.
This also uncovered and fixed two seperate issues with the patch:

• We need to use the XP_OHOS define, since OHOS is never defined by us
• The target_is_ohos check returning False still causes set_define to set the define (differing from the documentation), so we need to return None.

Fixes #582

[mukilan]

Would it make sense to define a constant for this domain value, maybe in the init.configure, rather than hard coding it?

[jschwe]

Yeah, that is definitely better than copying the same value everywhere.

[mukilan]

It looks like set_define also accepts a when argument. Is it not possible to use it here?

[jschwe]

I checked the implementation (set_define_impl) and the doc string says that when the value is false, the define will be explicitly undefined (-U).

[jschwe]

Ah, did you want to suggest to just remove the ohos_hilog_domain function and instead use when? that makes sense to me.

[mukilan]

Right, my idea was to see if we can just do set_define("OHOS_LOG_DOMAIN", "0xE0C4", when=target_is_ohos)

[jschwe]

I'm not really sure what is going on here:

@depends(target)
def target_is_ohos(target):
    return False


set_define("XP_OHOS", target_is_ohos)

This causes XP_OHOS to be defined.

@depends(target)
def target_is_ohos(target):
    return None


set_define("XP_OHOS", target_is_ohos)

This does not define XP_OHOS. This seems really weird to me, and goes against the documentation of set_define_impl and also how other defines are set.

[mukilan]

@jschwe is this PR still relevant?

[jschwe]

IMHO yes. I didn't proceed, because it seems I don't understand set_define and am not sure if I may be doing something wrong.

[mukilan]

This does not define XP_OHOS. This seems really weird to me, and goes against the documentation of set_define_impl and also how other defines are set.

I think the documentation of set_define_impl is just wrong, or its a bug in set_define_impl. The exisiting CROSS_COMPILE define is controlled by a boolean as well and I can see in my macOS builds it is defined to be an empty value in out/build/config.status which translates to -DCROSS_COMPILE='' . The same happens for XP_OHOS when False is returned from target_is_ohos.

The existing defines for XP_DARWIN, XP_WINDOW don't have a False return and instead let the predicate methods just return None by default, when the if doesn't succeed, which matches what you have in this PR.