DEPENDS is not a package specific variable

[michalsieron]

Moving conversation from #856 to a proper discussion

I noticed that DEPENDS:class-native += "qemu-system-native" in meta-oe/recipes-test/syzkaller/syzkaller_git.bb wasn't caught by check implemented in #856. It was caused by DEPENDS being listed as a package specific variable:

oelint-adv/oelint_adv/rule_base/rule_var_override_append.py

Line 20 in de2ffcd

'DEPENDS',

From the docs:

DEPENDS is a list of recipe names. Or, to be more precise, it is a list of PROVIDES names, which usually match recipe names. Putting a package name such as “foo-dev” in DEPENDS does not make sense.

For the same reason it doesn't make sense to make DEPENDS (not RDEPENDS) package specific.

[priv-kweihmann]

For the main title of this discussion, of course DEPENDS is not package specific, so some of the changes introduced in fe5bcce, do not make much sense.

Meaning

• DEPENDS
• DESCRIPTION
• SUMMARY
• SECTION

not entirely sure how they ended up there, but they can surely (or better to say, they need to be removed)

Patches welcome

[michalsieron]

Actually all other variables that you listed are package specific, but have a default value taken from non-specific definition.

SUMMARY and DESCRIPTION in meta/conf/bitbake.conf:

# The following two are commented out because they result in a recursive
# definition of the variable in some corner cases.  These are left in
# to illustrate the intended behavior.
#SUMMARY:${PN} ?= "${SUMMARY}"
#DESCRIPTION:${PN} ?= "${DESCRIPTION}"

and SECTION also from bitbake.conf:

$ grep -w 'SECTION' meta/conf/bitbake.conf
SECTION = "base"
SECTION:${PN}-doc = "doc"
SECTION:${PN}-dev = "devel"
SECTION:${PN}-staticdev = "devel"
SECTION:${PN}-dbg = "devel"
SECTION:${PN}-src = "devel"

[priv-kweihmann]

in that case the rule is actually a tiny bit faulty as it assumes that those contain a PN override.
IMO a patch would need to address a potential PN override and remove it but only if found

[michalsieron]

Hmm, TIL that quite a few more variables behave like that:

• ALLOW_EMPTY - meta/lib/oe/packagedata.py falls back to non package specific value
• CONFFILES - meta/lib/oe/package.py falls back to non package specific value
• DESCRIPTION - as explained in comment above
• INITSCRIPT_NAME - meta/classes-recipe/update-rc.d.bbclass expands it in path = d.expand("${D}${INIT_D_DIR}/${INITSCRIPT_NAME}"), because the package name is added to OVERRIDES
• INITSCRIPT_PARAMS - same as above
• INSANE_SKIP - uses package override for package_qa checks. Other checks use plain INSANE_SKIP (interestingly including 32bit-time, which is a package specific check)
• LICENSE - this can be set to be different for specific packages in recipe
• RDEPENDS - non package specific variant seems to be defined only for buildtools-tarball.bb, uninative-tarball.bb and in image.bbclass. It seems that package_(ipk|deb|rpm).bbclasses again simply add package name to overrides and resolve the variable values that way.
• RCONFLICTS, RSUGGESTS, RPROVIDES, RRECOMMENDS, REPLACES - I spotted RRECOMMENDS in image.bbclass, but in theory the rest could also be used in the same way
• SECTION - as explained in comment above
• SKIP_FILEDEPS - can be defined for entire recipe meta/lib/oe/package.py#L1587 or per package meta/lib/oe/package.py#L1599
• sstate - I am not sure what it is doing here. There is no such variable in use in oe-core?
• SUMMARY - as explained in comment above
• SYSTEMD_AUTO_ENABLE, SYSTEMD_SERVICE - uses get_package_var() to fallback to non package specific value

I think I checked all of currently defined in that self.pkgspecific list

[michalsieron]

@priv-kweihmann I would suggest merging #857 if you are happy with its current form and handling all that stuff above in a separate issue to keep it understandable. It might also be a good idea to see if the list can be somehow merged/reused in oelint.vars.pkgspecific?

[priv-kweihmann]

I will merge, if time allows tomorrow, but will stall the next release until a few more adjustments are made (also the improvement of error messages PR should go in before).

[michalsieron]

@priv-kweihmann sstate is also not a variable that is used anywhere in oe-core. But this allows me to guess why both it and DEPENDS ended up in the list:

$ git grep -rE '\w+:\$\{PN\}' | grep -E 'sstate|[^R]DEPENDS'
meta/classes-global/insane.bbclass:    # Some people mistakenly use DEPENDS:${PN} instead of DEPENDS and wonder
meta/classes-global/insane.bbclass:    if (d.getVar(d.expand('DEPENDS:${PN}'))):
meta/classes-global/insane.bbclass:        oe.qa.handle_error("pkgvarcheck", "recipe uses DEPENDS:${PN}, should use DEPENDS", d)
meta/classes-global/sstate.bbclass:SSTATE_PKGSPEC    = "sstate:${PN}:${PACKAGE_ARCH}${TARGET_VENDOR}-${TARGET_OS}:${PV}:${PR}:${SSTATE_PKGARCH}:${SSTATE_VERSION}:"
meta/classes-global/sstate.bbclass:SSTATE_SWSPEC     = "sstate:${PN}::${PV}:${PR}::${SSTATE_VERSION}:"

[priv-kweihmann]

that might really explains why it ended up in this list - but as this is a clear error, I'm happy to remove it

[michalsieron]

ok, I added its removal to #857