Re: using git to get a particular version of src

On Mon, Jan 25, 2021 at 12:35 PM Warner Losh <imp_at_bsdimp.com> wrote:

>
>
> On Mon, Jan 25, 2021 at 11:55 AM Warner Losh <imp_at_bsdimp.com> wrote:
>
>>
>>
>> On Mon, Jan 25, 2021 at 11:11 AM Yasuhiro Kimura <yasu_at_utahime.org>
>> wrote:
>>
>>> From: Chris <bsd-lists_at_bsdforge.com>
>>> Subject: Re: using git to get a particular version of src
>>> Date: Mon, 25 Jan 2021 09:21:44 -0800
>>>
>>> >> Hi,
>>> >> I have this version installed:
>>> >> 13.0-CURRENT #0 2ed50808d2b-c254384(main): Thu Nov 12 10:03:35 UTC
>>> >> 2020
>>> >> I'd like to get the sources for this (want to make a no-debug kernel)
>>> >> as I know
>>> >> this version works on this hardware.
>>> >> But -current has gone to 14 and what was -current is now
>>> >> 13-stable. What git
>>> >> incantation is required to get 2ed50808d2b-c254384 sources, given an
>>> >> empty
>>> >> /usr/src and git method of ssh://anongit_at_git.freebsd.org ?
>>> > I am by *no* means a GIT expert
>>> > but will
>>> > cd /usr/src
>>> > git up 2ed50808d2b
>>> > accomplish your intended task?
>>>
>>> Unfortunately it doesn't work as is expected in this case because hashes
>>> of src repostory changed on Dec 6 when it was still beta.
>>>
>>> HEADS UP: src hashes will respin/change this Sunday
>>> https://lists.freebsd.org/pipermail/freebsd-git/2020-December/000548.html
>>>
>>> In original case it was after this change so hash value can be used to
>>> checkout it. But this case is befere hash change. So there isn't hash
>>> 2ed50808d2b in current src repository any more.
>>>
>>> I also faced this problem when I tried to checkout source tree that
>>> corresponds to 20201119 snapshot of 13-CURRENT. Fortunately I still
>>> kept ISO image of it. So I did following steps to get the source tree.
>>>
>>> 1. Mount the ISO image
>>> 2. Extract src.txz in the ISO image to somewhere (e.g. /tmp/usr/src)
>>> 3. cd /usr
>>> 4. git clone https://git.freebsd.org/src.git
>>> 5. cd src
>>> 6. Checkout 65c207758a9 that was committed at Thu Nov 19 21:10:36
>>>    2020 +0000 (the last one committed on Nov 19, 2020)
>>> 7. diff -ru /tmp/usr/src /usr/src
>>> 8. If there are any differences, checkout f9fe7b28bc2 that is just
>>>    previous commit of 65c207758a9
>>> 9. Repeat step 7 and 8 until there is no difference between
>>>    /tmp/usr/src and /usr/src.
>>>
>>
>> We should see how hard it would be to convert c254384 into a git hash...
>>
>> So, for me:
>>
>> % git describe
>> vendor/tzdata/tzdata2020f-255971-gfa6662b3689e (so my tip of main is
>> 255971 vs 254384 or +1587)
>> % git log -1 main~1587
>> commit dda1987fe5dbf418b55195990896b0ef0a5b8e4a
>> Author: Mateusz Piotrowski <0mp_at_FreeBSD.org>
>> Date:   Tue Nov 17 12:04:29 2020 +0000
>>
>>     Add an example for the -s flag
>>
>> which is a little late. If you checked out the source and built it ASAP,
>> then I'd suggest:
>>
>> commit f14436adc61a52b29d035791c0b90c0b221bea9a
>> Author: Hans Petter Selasky <hselasky_at_FreeBSD.org>
>> Date:   Thu Nov 12 09:26:01 2020 +0000
>>
>>     Add a tunable sysctl, hw.usb.uaudio.handle_hid, to allow disabling the
>>     the HID volume keys support in the USB audio driver.
>>
>> which is the version just before that. Or if you installed from a
>> snapshot and rebuilt, The 20201112 snapshot was likely built from
>>
>> commit 38033780a3f4ed616d638fc0e9ef9a4d309f1fb3
>> Author: Kyle Evans <kevans_at_FreeBSD.org>
>> Date:   Wed Nov 11 22:35:23 2020 +0000
>>
>>     umtx: drop incorrect timespec32 definition
>>
>
> Oh! git describe is the wrong thing. We're using git rev-list --count, so:
>
> % git rev-list --count main
> 256115
> %  echo $((256115-254384)
> 1731
> % git log -1 main~1731
> git log -1 main~1731
> commit 94275e3e698b223ccc42e3a637b6667216ca6360
> Author: Mateusz Guzik <mjg_at_FreeBSD.org>
> Date:   Tue Nov 10 01:31:06 2020 +0000
>
>     threads: remove the unused TID_BUFFER_SIZE macro
>
> But, there's a problem, or a disconnect. git rev-list counts both the
> commits on main, and the number of commits for merge commits that are on
> the second parent back to the last common ancestor. this is surprising
> behavior, and we'll likely change it, but for systems before the change you
> need to do some searching:
>
> % git rev-list --count main~1731
> 254360
> % git rev-list --count main~1707
> 254384
>
> So this is the one you really want:
>
> commit 26007fe37c06fe0b61634083befb7f9eb87c08c0
> Author: Mateusz Guzik <mjg_at_FreeBSD.org>
> Date:   Wed Nov 11 08:51:04 2020 +0000
>
>     thread: add more fine-grained tidhash locking
>
> (that's assuming the commit counts from before the re-hashing were
> identical, which I have no way of knowing or testing).
>
> The truth is, though, any of these will likely be 'good enough' to have a
> working system.
>

P.S. Due to this mismatch, I'm going to propose we move to git rev-list
--first-parent  --count

Warner