On Tue, Sep 03, 2019 at 10:04:42AM +0200, Wolfgang Denk wrote:
Dear Tom,
In message a78f0b04-c3f7-45d5-e9ac-90522dbefc2e@denx.de Heiko Schocher wrote:
I am just testing U-Boot Environment flags and they do not work anymore with current mainline U-Boot ...
...
reason is your commit:
commit 7d4776545b0f8a8827e5d061206faf61c9ba6ea9 Author: Patrick Delaunay patrick.delaunay@st.com Date: Thu Apr 18 17:32:49 2019 +0200
env: solve compilation error in SPLLooking into the history of this, I wonder if we could / should have prevented this.
Looking over my scripts, yes, I overlooked the problem. The 'edison' platform shows the same issue that Heiko's platform does but I overlooked the size change. I'm modifying my script currently so it will show more details and this should jump out more rather than the size noise of "changes in a general area". Now interesting enough, sandbox didn't blow up here but does also enable the env flags options.
As far as I can see, Patrick's patch series has not been reviewed by others, probably because general intetest in STM32 is not that big at the moment. I can see no Acked-by:, Reviewed-by: nor Tested-by: tags - nothing.
The whole patch series was then pulled from the u-boot-stm repository.
However, there was not only STM related code in there. There were changes to common code like the environment handling. common code was changed without review and without testing.
To be clear, it was tested, but sadly the environment flags code is not heavily used / enabled. More in a moment.
Are there ways to prevent this?
Yes, we can appeal to the custodians to be more careful, but I assume they are already doing their best.
It might have even been better if this had been a sub-system with a clear maintainer, but there is no such person for the environment code.
How can we prevent this in the future?
Should we define "interested developers" for such areas that have no custodian (the "Designated reviewer") entry in the MAINTAINERS file could be used for this, for example)?
This, along with some other environment related patches were things I was keeping an eye on to see if perhaps Joe would have had time to look at before it went in (as the env flag stuff came from him). I also try and make use of the "Needs Review / ACK" flag in patchwork for things that stand out. Looking over the merge contents again, that particular one would not have.
So, things that would help in the future: - An explicit environment maintainer - test.py tests for the environment flags, but only if they're also run on some platform(s) that also would have failed here. Perhaps we need to enable more functionality in something like qemu-x86 that is less of a special case build than sandbox is? In fact, since I know we have the QEMU targets in for "real" uses and not just testing, and while I worry about adding in more complex logic, we might want to rework the "build and run test.py in QEMU" parts of CI to first make use of scripts/kconfig/merge_config.sh to turn ON a whole bunch of testing related options.