Very good explanation. It clarifies something. I read some comments about 
using "subtree" vs "submodules". In most places people give a bad score to 

Четвер, 25 вересня 2014 р. 17:05:21 UTC+3 користувач Konstantin Khomoutov 
> On Thu, 25 Sep 2014 04:28:16 -0700 (PDT) 
> Віктор Невідомий < <javascript:>> wrote: 
> > > you have to turn each of these directories into separate 
> > > repositories and either use the so-called "subtree merging" or 
> > > submodules. 
> > 
> > I was going to do so. Maybe it was not clear from my scheme in 
> > original question. I edited it to clarify. 
> > 
> > Proj1_2_common/ 
> >      .git/ 
> >      ... 
> > Proj1_2_3_common/ 
> >      .git/ 
> >      ... 
> > Proj1/ 
> >      .git/ 
> > 
> > And after this again: what is advantages of "subtree" or "submodules" 
> > over just add "Proj1_2_common/" to project in IDE and use it repo 
> > separately? 
> Ah, I see now, thanks. 
> The problem with simply adding them all into an IDE project is that 
> tracking that project's history with Git (I mean, tracking the files 
> comprising what constitutes a project in your IDE, such as .sln and a 
> set of *.csproj files for an C#/.NET application, and may be also some 
> code files etc) will produce a series of commits which, themselves, 
> contain no record of which exact states all of the referenced 
> subprojects were in when that commit has been recorded. 
> Let me try to explain that in more words. 
> Suppose you did what you intended, and just slapped a bunch of 
> git-clone'd projects under a single directory, and added references 
> to the files in them to your IDE's project.  So far so good. 
> Now some time passes and some of those referenced projects get updated. 
> You'll typically `cd` into each of the referenced projects and do 
> `git pull` (or may be something more appropriate) there -- to bring the 
> latest changes in.  You will then possibly make some adjustments to your 
> "superproject" and commit these changes. 
> Now you see that should you have the need to check out some *past* 
> revision of your superproject (maybe during `git bisect` or to just 
> fork a branch off some prior state etc), you'll discover that the 
> commit you're about to check out has no idea about which precise states 
> of the subprojects it references have been checked out when that commit 
> has been recorded.  That happens because the synthetic state of all the 
> checked out projects was "ad hoc", and was never recorded anywhere, 
> anyhow. 
> Enter subtree merging or submodules. 
> With subtree merging, you have histories of subprojects recorded in 
> your repository.  You merge (and later re-merge) their new state 
> into your superproject from time to time, and hence any commit you 
> record "on the top level" -- for the files comprising the superproject 
> itself -- automatically references the correct states of all the 
> subprojects -- because they're in the same repository. 
> With submodules, your superproject maintains a list of submodules, 
> and each commit recorded in the superproject records SHA-1 names 
> of the commits currently checked out in each submodule at the time 
> the commit is created. 
> Hence, with either approach, when you later check any of your past 
> revisions of the superproject, the exact state of the whole thing is 
> reconstructed. 
> Pros and cons of these approaches are: 
> Subtree merging has everything in the single repository: 
> easier to carry around and view the history. 
> But this comes at the cost of having the histories 
> of the subproject in the superproject's repository. 
> Submodules require accessing other repos when you clone 
> the superproject and hence the superproject's repo is not 
> free-standing.  On the other hand, there is no history duplication. 
> > And what about history of old and new repos? 
> Either approach will make histories of subprojects available 
> when working on the superproject, though via different means. 
> Of course, you will have a single point in the history of your 
> superproject where you will have started using either of the 
> approaches explained above.  If you want to somehow retrofit past 
> states of the subproject's histories intertwined with certain past 
> states of the superproject this is another task completely and, 
> while supposedly doable, this will be hard and tedious and manual 
> to get done. 
> > it still would be better if you have copied the original answer inline 
> > Only answer without question? 
> My bad: I meant question. 

You received this message because you are subscribed to the Google Groups "Git 
for human beings" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
For more options, visit

Reply via email to