Skip to contentSkip to conversation

git push vs. slow connection times - local commit resolution is too late

DISCUSSION Eric Blake 3 participants 4 messages Jun 15, 2016

lore
Share & tools
lore

Eric Blake

author
#0 lore raw

I've been bitten several times by this bug now, so I'll report it in the hopes that someone knows what to do to patch it.

Scenario - I have a remote repository that takes on the order of 10 seconds to connect to, for any operation like 'git push'. I know that there is a lag, so I intentionally have two terminals open, both visiting the same directory, one for interaction with the remote, and the other for acting on the local repository, with the hope that I can do useful work in the second terminal rather than idly waiting on the lag in the first terminal.

In the middle of rebasing a patch series, where I want to incrementally push the patches that I have gotten through so far, I used the following steps:

On the remote-interaction terminal, I push the current state of my tree:
git push remote HEAD:master

On the local-interaction terminal, I move on to the next patch:
git rebase --continue

then, to my horror, I find out that the commit I'm working on locally has already been pushed! Why? Because 'git push remote HEAD:master' does not determine which commit 'HEAD' refers to until _after_ it has established a connection to remote, but the stupid 10-second lag was long enough that my actions in my second terminal have changed HEAD in the meantime.

I would really love it if 'git push' would resolve all local references _prior_ to trying to connect to the remote server, rather than waiting until after the connection. That way, my remote-interaction terminal can truly be a type-it-and-forget-it terminal, where the push action I requested reflects the state of the tree at the time I requested it, rather than picking up changes made later in another terminal.

Signature · 2 lines
Eric Blake   [email protected]    +1-801-349-2682
Libvirt virtualization library http://libvirt.org
#1 lore raw
Quoted 28 lines · Eric Blake

I've been bitten several times by this bug now, so I'll report it in the hopes that someone knows what to do to patch it.

Scenario - I have a remote repository that takes on the order of 10 seconds to connect to, for any operation like 'git push'. I know that there is a lag, so I intentionally have two terminals open, both visiting the same directory, one for interaction with the remote, and the other for acting on the local repository, with the hope that I can do useful work in the second terminal rather than idly waiting on the lag in the first terminal.

In the middle of rebasing a patch series, where I want to incrementally push the patches that I have gotten through so far, I used the following steps:

On the remote-interaction terminal, I push the current state of my tree:
git push remote HEAD:master

On the local-interaction terminal, I move on to the next patch:
git rebase --continue

then, to my horror, I find out that the commit I'm working on locally has already been pushed! Why? Because 'git push remote HEAD:master' does not determine which commit 'HEAD' refers to until _after_ it has established a connection to remote, but the stupid 10-second lag was long enough that my actions in my second terminal have changed HEAD in the meantime.

Quoted 6 lines

I would really love it if 'git push' would resolve all local references _prior_ to trying to connect to the remote server, rather than waiting until after the connection. That way, my remote-interaction terminal can truly be a type-it-and-forget-it terminal, where the push action I requested reflects the state of the tree at the time I requested it, rather than picking up changes made later in another terminal.

I agree with your suggestion. But as a quick fix, can you do this?
git push remote $(cat .git/HEAD):master

Damned inconvenient, though.

Phil

#2 lore raw
Quoted 2 lines · Phil Hord

I agree with your suggestion. But as a quick fix, can you do this?
git push remote $(cat .git/HEAD):master

You mean $(git rev-parse HEAD), don't you?

Andreas.

Signature · 3 lines
Andreas Schwab, [email protected]
GPG Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#3 lore raw

Andreas Schwab helpfully chimed in

Quoted 6 lines
Quoted 2 lines · Phil Hord

I agree with your suggestion. But as a quick fix, can you do this?
git push remote $(cat .git/HEAD):master

You mean $(git rev-parse HEAD), don't you?

Yes, I do. :-) But I don't use it enough to remember it, and the man pages are not cross-referenced well enough for me. So I shrugged.

Thanks for helping me out.

Phil