Hacker News new | past | comments | ask | show | jobs | submit login

That's one of the things I hate about git (I like git a lot, in general). It makes it too easy to make irreversible damage.

"git clean", "git reset --hard", "git checkout" all provide ways to irreversibly trash the working tree.

"git clean" has no other function except doing irreversible deletions, so that's OK, IMO.

"git reset --hard" is a pretty useful way to make a branch jump around between places, so having a secondary feature of trashing the working tree is unacceptable, IMO.

"git checkout" is a useful way to take a file from a specific commit. The fact it may also trash uncommitted changes is also a secondary feature. I think without "--force", "checkout" should refuse to irreversibly lose content.




"git clean" does nothing without -f, or the user changing clean.RequireForce=false

"git reset" does nothing irrevocable unless you add --hard (why should you also need to add --force?)

"git checkout" will only trash uncommitted changes if you ask it to specifically. (e.g. 'git checkout <file to overwrite>').

side note: using 'git reset --hard' to teleport a branch around is bad form, imho.


If you take a look at the man pages for all of those commands you'll see they explicitly mention that your working tree will be affected. They are written explicitly to do that. 'git reset' can be called with '--mixed' or '--keep' which might be more along the lines of what you're looking for. If you're ever unsure about the result of a given command make a copy of your current working tree before applying it. I like git because it explicitly doesn't hold your hand through things like this. It expects you to know what you're about to do.


> If you take a look at the man pages for all of those commands you'll see they explicitly mention that your working tree will be affected.

I'm not complaining that these commands mismatch the spec. I'm complaining that the spec for these commands is badly designed.

> They are written explicitly to do that

I agree about git clean, but git reset --hard and git checkout do more than just changing the working tree.

> If you're ever unsure about the result of a given command make a copy of your current working tree before applying it. I like git because it explicitly doesn't hold your hand through things like this. It expects you to know what you're about to do.

You misunderstood my problem. I know exactly what these commands do.

I consider "reverting my changes in the working tree" to be a separate and irreversible thing of whatever "git reset" and "git checkout" do, and want git to ask for approval before accidentally doing irreversible damage.

The main thing protecting me from accidental errors/deletions/etc is my revision control system. But in this case, the irreversible damage is done BY the revision control system.

Here are a few examples of what I mean:

* I jump to the wrong shell. In it, I have an old "git status" that says my working tree is clean, but at the wrong location. I forget that I already moved it, and started working on a feature. I use "git reset --hard <correct-place>" and without warning, I also get my working tree changes irreversibly lost without any question.

* I press <up> a few times in my shell, to reach a previous command, hit <return>. OOPS, that command was not the one I intended. Instead it was a "git reset --hard <someplace>" or perhaps a "git checkout"!

* I may execute a command like "git checkout <refspec> <some_path>" during a merge in order to pick some version of that file. If I accidentally press <return> too early, I may lose all of my working tree changes in the directory.

I don't actually lose my working tree changes with git, because I'm aware of these issues, so I am very careful about which commands I replay.

But it would be nice to know that since I have a revision control backup of my files, I don't need to be as careful, as long as I don't use "clean", "rm" and other commands meant to lose information.

P.S: It would be nice if "checkout" stored a stash or such of any overwritten content so it isn't irreversible unless given a flag not to.

tl;dr: Good UI design dictates that user operations should be reversible whenever possible. Irreversible operations should require explicit confirmation (or at least "--force") and ideally should have no other effect except the specific irreversibility.


While all of what you say rings true - it's been said before and bears repeating - git isn't a version control system as much as it is a set of version control tools from which you can design a workflow that meets your needs.


That's a poor excuse for bad defaults.

The canonical way to use git is by the git syntax. Many (most?) people don't build scripts and aliases around it. It works just fine when used raw.

However, I'm with Peaker in that losing your working copy is enough of a nuisance to deserve an interactive prompt, just like 'rm' prompts by default. Confident users could always disable the prompt with an environment-variable or .gitconfig flag.


'rm' prompts by default? since when?

Btw, which defaults are, in your opinion, "bad"?


In my experience, Mercurial is a little bit more friendly in this regard. You usually have to add a force argument for it to do anything that can cause irreversible damage.

Which can be frustrating as well, at least with git the commands work the first time :)

A better approach (as long as no external pushing/pulling is done) IMO would be to execute the command immediately, but allow undo functionality.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: