On Wed, 18 Dec 2024, Mark Millard wrote:
On Dec 18, 2024, at 10:55, Bjoern A. Zeeb <b...@freebsd.org> wrote:
On Wed, 18 Dec 2024, Mark Millard wrote:
On Dec 18, 2024, at 10:12, Bjoern A. Zeeb <b...@freebsd.org> wrote:
On Wed, 18 Dec 2024, Mark Millard wrote:
Hi Mark, ports@,
As an example context, I am referencing:
https://pkg-status.freebsd.org/ampere2/build.html?mastername=main-armv7-default&build=peb87cb7f3aa2_s48d92db080
which is showing wifi-firmware-kmod-release-20241017 2 times in Ignored ports:
# Package Origin Skipped Reason
1623 wifi-firmware-kmod-release-20241017 net/wifi-firmware-kmod@release 0
Dependent port net/wifi-firmware-iwlwifi-kmod@8000 |
wifi-firmware-iwlwifi-kmod-8000-20241017.1500029_1 ignored
1629 wifi-firmware-kmod-release-20241017 net/wifi-firmware-kmod@release 0
Dependent port net/wifi-firmware-iwlwifi-kmod@22000 |
wifi-firmware-iwlwifi-kmod-22000-20241017.1500029_1 ignored
Funny. I'd expect it to try all or just the furst dependency and if
that gets ingnored then to also ignore the rest right away.
Parallel built problems in pourdierer?
I'll use the log file context to try a different wording to try to make
clear what I'm reporting. I'm also CC'ing Bryan Drewery so he can cross
check if I'm just wrong about something.
I'll note that here is only one:
https://pkg-status.freebsd.org/ampere2/data/main-armv7-default/peb87cb7f3aa2_s48d92db080/logs/wifi-firmware-kmod-release-20241017.log
and it is for the net/wifi-firmware-iwlwifi-kmod@22000 when looking at its
content. I appears that log file content replaced the earlier content for
net/wifi-firmware-iwlwifi-kmod@8000 (because of the lack of a unique log file
name).
Are you saying "unique log file name" here as the port name is distinct
with the name of the flavor?
As I understand things, package names for a flavor are supposed to indicate the
flavor
explicitly in some way: such a one-to-one correspondence is required as I
understand
things.
net/wifi-firmware-iwlwifi-kmod@8000 is using the package name:
wifi-firmware-kmod-release-20241017
also:
net/wifi-firmware-iwlwifi-kmod@22000 is using the package name:
wifi-firmware-kmod-release-20241017
The 2 flavors have the same package name.
No. They are both child packages of the metport pacakge.
The metaport has two flavours by the names:
wifi-firmware-kmod-NNNNNNNN
wifi-firmware-kmod-release-NNNNNNNN
The metaport package depends on ports with multiple flavours.
Have a look here:
https://cgit.freebsd.org/ports/tree/net/wifi-firmware-kmod/Makefile
So for example the port:
net/wifi-firmware-iwlwifi-kmod
has flavours:
7000
8000
9000
22000
ax210
bz
which then individually expand to package names like:
wifi-firmware-iwlwifi-kmod-FLAVOUR-MMMMMMMMM
with FLAVOUR being one of the above and MMMMMMMM being the version of the
iwlwifi firmware. And then there is a verions without the -FLAVOUR in
the middle which contains all of the iwlifi firmware which is the
default target for net/wifi-firmware-iwlwifi-kmod -- which the metaport
does not depend on so you are not seeing it there.
Same goes for rtw88, rtw89, ath1[012]k, mt76.
So it's a tree you are looking at.
Very much like the gpu-firmware port for example.
https://docs.freebsd.org/en/books/porters-handbook/flavors/
reports:
QUOTE
Example 4. Flavor Specific PKGNAME
As all packages must have a different package name, flavors must change theirs,
using flavor_PKGNAMEPREFIX and flavor_PKGNAMESUFFIX makes this easy:
FLAVORS= normal lite
lite_PKGNAMESUFFIX= -lite
END QUOTE
Note the "must". The combination:
net/wifi-firmware-iwlwifi-kmod@8000 is using the package name:
wifi-firmware-kmod-release-20241017
also:
net/wifi-firmware-iwlwifi-kmod@22000 is using the package name:
wifi-firmware-kmod-release-20241017
This is via:
BASEDIR= ${.CURDIR}/../wifi-firmware-kmod
. . .
.include "${BASEDIR}/Makefile.inc"
That in turn uses:
No! Makefile.inc != Makefile
PORTNAME= wifi-firmware-kmod
. . .
FLAVORS= default release
release_PKGNAMESUFFIX= -release
that ignores the @8000 and @2200 but can not validly do so under the wording.
Poudriere itself is reporting the actual pack name in use for
both is: wifi-firmware-kmod-release-20241017
I think we just need to wait for Bryan D. to comment about which of us
is correct for the interpretation.
I will say that if you are correct, poudriere depends on using package
names inappropriately, such as that destruction of the log file for
one flavor by the log file for another flavor that uses the same package
name (since only the package naming was used in forming the log file name).
An example of the consequences:
Go try to find a log file in:
https://pkg-status.freebsd.org/ampere2/data/main-armv7-default/peb87cb7f3aa2_s48d92db080/logs/
for the check of:
net/wifi-firmware-iwlwifi-kmod@8000
There is not one. That is because the package name does not indicate the flavor
at all and that in turn means that the log file name does not contain a
reference
to the flavor name and the second check replace's the first check's file.
As I understand it, poudriere's infrastructure does not support having multiple
flavors of a port contributing to the same package for the port (by name).
. . .
Please ignore the Skipped ports part of my report.
Those dependency packages for wifi-firmware-kmod-NNNNNNNN respectively
wifi-firmware-kmod-release-NNNNNNNN are not built for armv7:
Makefile.inc:ONLY_FOR_ARCHS= aarch64 amd64 i386
so hopefully the "parent" (metaport) gets skipped as well once the first
one "fails" with the IGNORE.
If pourdierer does not fully grasp it, so be it. The port names are
unique with the two flavours there are within the ports framework.
I'd expect you'd have to file a pourdriererer bug then?
Otherwise if I get this entirely wrong, can you please explain it more
precisely?
r15:7
===
Mark Millard
marklmi at yahoo.com
--
Bjoern A. Zeeb r15:7
