On Mon, Jun 13, 2022 at 06:39:19PM +0300, Andy Shevchenko wrote:
> > Do a `git --grep=revert`. Some of them you can grep for "This reverts
> > commit 8bdc2a190105e862dfe7a4033f2fd385b7e58ae8." but there are a lot
> > which are not machine parsable
>
> Why not? The format of the string hasn't been changed, no difference from other
> patterns.
>
With the Fixes tag you can just do a:
git log | grep Fixes: | cut -d : -f 2 | cut -d '(' -f 1
It's easily machine parseable. But if you look at the examples I posted
they're stuff like this:
This reverts commit 9eec1d897139e5d ("squashfs: provide backing_dev_info
It can't be grepped for, it needs a human to try figure it out. And the
reason for that is that we always tell people that git hashes need to
be in a specific format which git revert violates.
Having two hashes *is* duplicative but if we're to delete a hash we
should do Hans did and delete the "This reverts commit
fb561bf9abde49f7e00fdbf9ed2ccf2d86cac8ee." line. (As an aside, in
that commit the reverts line is not a Fixes line. The original commit
was a temporary hack and it was deleted when it was no longer required.
So reverts and Fixes are not the same. Reverts is ambiguous.)
The problem with the reverts line is that most other people besides Greg
only look for the Fixes tags. It had never occured to me to look for
the reverts line. I was just reading an LWN article about bugs in
-stable and LWN only used Fixes tags, not reverts lines. Or when people
are backporting patches I tell them to look for the Fixes tags to see if
they are backporting buggy patches. If they're searching
lore.kernel.org most people will use the 12 char git hash instead of the
full hash.
My main problem with `git revert` is that it writes the commit message
for you and it does it really badly. When I'm reviewing those patches
I have to tell people, "No, never use git revert format. Send normal
patches." I always tell them to redo it like Hans did.
Subject is wrong:
https://lore.kernel.org/all/20220614011528.32118-1-tangmeng@xxxxxxxxxxxxx/
No Signed-off-by:
https://lore.kernel.org/all/BN9PR12MB5257FB6CA192626D5D108C2DFCAB9@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx/
Terrible commit message:
https://lore.kernel.org/all/20210414233533.24012-2-qingqing.zhuo@xxxxxxx/
No commit message.
https://lore.kernel.org/all/20220613132116.2021055-2-idosch@xxxxxxxxxx/
These are just the first view I looked at from yesterday afternoon.
Almost every patch with Revert in the subject needs to be NAKed.
regards,
dan carpenter
