gitup

… Cloning/pulling the source can take up to 2GB of memory. …

Is this the likeliest cause of "Killed", where the OS has no swap?

1629567620517.png
 
Just gave gitup a first try and started playing. I started with:

Code:
{
  "defaults" : {
    "proxy_host"        : "proxy",
    "proxy_port"        : "3128",
    "port"              : 443,
    "host"              : "git.freebsd.org",
    "low_memory"        : false,
    "display_depth"     : 0,
    "verbosity"         : 1,
    "work_directory"    : "/tmp/gu",
  },

  "ports" : {
    "repository_path"  : "/ports.git",
    "branch"           : "main",
    "target_directory" : "/tmp/g",
    "ignores"          : [ "INDEX-11", "INDEX-12", "INDEX-13", "INDEX-14" ],
  }

}

and non-existing dirs "/tmp/g" and "/tmp/gu". Now I run "gitup ports" and it fills "/tmp/g" with the
current ports-tree. Out of curiosity I now run "gitup -r ports" and stuff in "/tmp/g" gets deleted.
Either I didn't get the logic or this is somewhat buggy :).
 
😩 Looks like I'll stick with portsnap until I see it no longer working. Or I might learn straight git (not a priority for me, got other stuff on my plate, I just like the idea) 😋
 
Not much to learn. "git clone" followed by regular "git fetch", "git diff" (to see what would be done) and "git merge".
Can be combined into "git pull" but I like to do the single steps.

However, this gives you a bloated ports directory. Mine is 2,2GB with 50% sitting in .git. You will laugh but I'm
still a ctm-user (since 25+ years). But ports is no more delivered via ctm and Julian has no time to fix it. So
I was happy to discover gitup which seems small but maybe not that mature yet :).

Another thing is that I have a few local modifications in my ports tree (stuff that can't be addresse via EXTRA_PATCH_TREE)
and I used to have a wrapper which restores the original files, applies the ctm files, and reapplies my own mods -- telling me
if their original versions were touched.

I have adopted this wrapper for the above git commands but I'd prefer to use something like gitup.
 
Now, what I'd like to have is an update mode where gitup leaves unknown files alone.
E.g., something like "ignores" for everything that does not get deleted or modified
actively from upstream. An example:

1. I run "gitup ports" to check out all for the first time

2. Now I add "/usr/ports/x/y/myfile"
3. Now I modify "/usr/ports/a/b/Makefile"

4. Someone modifies "/usr/ports/a/b/Makefile" upstream
5. Someone deletes"/usr/ports/c/p/distfile" upstream

6. Now I run "gitup ports" again. Normally it would now modify "Makefile",
delete "distfile" and "myfile". But I would like to have "myfile" left alone...
 
Because gitup will not let mentioned files alone. So I will have to live with git (and its bloat)
or hack gitup to do what I want. My hope was that gitup can be used as is for that but apparently
this is not the case.
 
Yep. Bloat was about 25% smaller but growing. I am happy now with the modified gitup ;-).
Not sure what you're talking about.
Code:
$ git clone --depth 1 -b 2021Q4 https://git.FreeBSD.org/ports.git
Cloning into 'ports'...
remote: Enumerating objects: 178456, done.
remote: Counting objects: 100% (178456/178456), done.
remote: Compressing objects: 100% (167126/167126), done.
remote: Total 178456 (delta 9323), reused 118397 (delta 6961), pack-reused 0
Receiving objects: 100% (178456/178456), 73.81 MiB | 9.18 MiB/s, done.
Resolving deltas: 100% (9323/9323), done.
Updating files: 100% (142020/142020), done.
$ du -sh ports                                                                          
1.0G    ports
$ cd ports
$ du -sh .git                                                                     
 94M    .git
 
Make sure your /usr/ports isn't filling up with all the work/ directories.
 
Gitup gives you one branch at a time. Let's compare apples to apples. Not sure what doing it for 20 days would prove.
Yep, you suggested git shallow clones and I commented on it. To clarify:
I tried git shallow clones for approx. 20 days, syncing one or two times a day.
 
Back
Top