[OE-core] SDK and external toolchain

[Mark Hatle]

On 7/26/12 1:14 PM, Chris Larson wrote:
> On Thu, Jul 26, 2012 at 9:37 AM, Mark Hatle <mark.hatle at windriver.com> wrote:
>>> index 44284c3..f5fd4d7 100644
>>> --- a/meta/classes/toolchain-scripts.bbclass
>>> +++ b/meta/classes/toolchain-scripts.bbclass
>>> @@ -136,7 +136,7 @@ toolchain_create_sdk_env_script_for_installer () {
>>>    #we get the cached site config in the runtime
>>>    TOOLCHAIN_CONFIGSITE_NOCACHE = "${@siteinfo_get_files(d, True)}"
>>>    TOOLCHAIN_CONFIGSITE_SYSROOTCACHE =
>>> "${STAGING_DATADIR}/${TARGET_SYS}_config_site.d"
>>> -TOOLCHAIN_NEED_CONFIGSITE_CACHE = "${TCLIBC} ncurses"
>>> +TOOLCHAIN_NEED_CONFIGSITE_CACHE = "ncurses"
>>
>>
>> That is incorrect.. the CONFIGSITE_CACHE should be generated for the TCLIBC.
>> If you don't do that, then you will be running the same configure steps --
>> looking for basic glibc information over and over and over, causing a fairly
>> expensive performance penalty.
>
> No, he's right, this is a bug in toolchain-scripts.bbclass. We could
> work around it in the recipe via RPROVIDES_${PN}, but
> TOOLCHAIN_NEED_CONFIGSITE_CACHE feeds directly into task 'depends', so
> it's pulling in ${TCLIBC} explicitly rather than the more accurate
> 'virtual/libc'.
>

The config site file though isn't called "virtual/libc".  My understanding what 
that all that variable did was set the list of config site files, but didn't 
directly affect the dependency mapping.  If it does, then there is a definite 
issue w/ dependency vs file mappings.

--Mark