# The Universe of Discourse

Wed, 24 Nov 2010

git-reset
The Git subcommand git-reset is very frequently used, and is one of very few commonly-used Git commands that can permanently destroy real work. Once work is in the repository, it is almost completely safe from any catastrophe. But git-reset also affects the working tree, and it is quite possible to utterly destroy a day's work by doing git-reset --hard at the wrong time. Unfortunately, the manual is unusually bad, with a huge pile of this stuff:


----------------------------------------------------
A       B     C    D     --soft   A       B     D
--mixed  A       D     D
--hard   D       D     D
--merge (disallowed)

----------------------------------------------------
A       B     C    C     --soft   A       B     C
--mixed  A       C     C
--hard   C       C     C
--merge (disallowed)



Six more of these tables follow, giving the impression that git-reset is quite complicated. Sure, I'm gonna memorize 256 table entries. Or look up the results on the table before every git-reset. Not.

The thing to notice about the two tables I quoted above is that they are redundant, because the second one is simply a special case of the first, with D replaced by C. So if you were really in love with the tables, you might abbreviate the 64 table entries to 28:

           working index target         working index HEAD
----------------------------------------------------
A       B    C      --soft   A       B     C
--mixed  A       C     C
--hard   C       C     C
--merge (disallowed)

But even this is much more complicated than it should be. git-reset does up to three things:
1. It points the HEAD ref at a new 'target' commit, if you specified one.
2. Then it copies the tree of the HEAD commit to the index, unless you said --soft.
3. Finally, it copies the contents of the index to the working tree, if you said --hard.
If you compare this with the table above, that is what you will see. The three points above replace at least 60% of the tables. Most of the rest concerns the less-frequently used --merge and --keep options and the circumstances in which the tree is considered to be in "good order".

Tables are good for computers to understand, because they have a uniform format and computers are unfazed by giant masses of redundant data. The computer will not understand the data regardless of how well-structured they are, so there is no reason to adopt a representation that showcases the structure.

For humans, however, tables are most useful when there is no deeper understanding of the structure to be had, because the structure tends to get lost in the profusion of data, as it did here.

[ Thanks to Aristotle Pagaltzis for pointing out that git checkout can also destroy the working tree, and for other corrections. ]