Rebase without re-checkout

TL;DR: when using multiple branches, prefer to use “git rebase –onto origin/master origin/master ${branch}” if you want to checkout and then rebase a branch.

Let’s say you have the working history:

       /-*-* A
*-*-*-*-*-*-* origin/master
        \-*-*-* B

and you want to push the branch A because your CL has been approved. You have to rebase A on origin/master, then build and run test and do git cl push. Once this is done, you now have the following history:

*-*-*-*-*-*-*-@ origin/master (after push)
        \-*-*-* B

Now you want to go back to work on B. Until recently, I used the following commands (striked-ed out as I now consider them broken):

$ git checkout B
$ git rebase origin/master

This has the disadvantage of marking as dirty all the files that are in the delta between the point B diverged from origin/master in addition to all the files modified in B since the branch was created.

There is however a better way:

$ git rebase –onto origin/master origin/master B

This will still rebase B on top of origin/master, but will not checkout B first, so only the files that have been modified in B will be touched and considered dirty by the build. Depending on how large the change in origin/master have been, this can be the difference between rebuilding thousand of files and rebuilding just a few tens.

How to print Chromium string16 values from lldb using python scripts

I found this here: with a minor tweak of if -> elif

Thanks stack overflow!

Python file:

$ cat ~/work/lldbscripts/
import lldb
def unsignedshortptr_SummaryProvider(valobj, dict):
  e = lldb.SBError()
  s = u'"'
  if valobj.GetValue() != 0:
    i = 0
    newchar = -1
    while newchar != 0:
          # read next wchar character out of memory
      data_val = valobj.GetPointeeData(i, 1)
      size = data_val.GetByteSize()
      if size == 1:
        newchar = data_val.GetUnsignedInt8(e, 0)    # utf-8
      elif size == 2:
        newchar = data_val.GetUnsignedInt16(e, 0)   # utf-16
      elif size == 4:
        newchar = data_val.GetUnsignedInt32(e, 0)   # utf-32
        return "<errorsize: %i>" % size
        return '<errorfail>'
      i = i + 1
      # add the character to our string 's'
      if newchar != 0:
        s = s + unichr(newchar)
  s = s + u'"'
  return s.encode('utf-8')

init file:

$ cat .lldbinit
command script import /Users/justincohen/work/lldbscripts/
type summary add -F unsignedshortptrsummary.unsignedshortptr_SummaryProvider "unsigned short *"

Gives you:

(lldb) p new_text
(string16) $5 = {
  (std::basic_string<unsigned short, base::string16_char_traits, std::allocator<unsigned short> >::_Alloc_hider) _M_dataplus = {
    (unsigned short *) _M_p = 0x110625cc "This is a string16!!"

Git clean. No seriously, clean it.

Working on Chrome, there are lots of dependencies that get pulled, and when I make untracked temporary changes in those dependancies sometimes I just want everything back to normal. So, here’s the scorched earth method:

git clean -f -f -x -d

This blows away everything git doesn’t know about. Then I do another clean dependency pull and I’m back to the beginning!

mach.h, dumping vm info

#import "mach/mach.h"
- (void)logTimer {
  task_basic_info info;
  mach_msg_type_number_t size = sizeof(info);
  kern_return_t kerr = task_info(mach_task_self(),
  if ( kerr == KERN_SUCCESS ) {
    NSLog(@"task_info: %d.2MB\n", info.resident_size / 1048576);
  else {
    NSLog(@"task_info failed with error %d ( 0x%08d ), '%s'\n",
        kerr, kerr, mach_error_string(kerr));

Another weird git problem

Lately whenever I have to rebase I get a weird error about how I have a dirty tree and need to commit files.

Falling back to patching base and 3-way merge...
error: Your local changes to the following files would be overwritten by merge:
Please, commit your changes or stash them before you can merge.
Failed to merge in the changes.

Well this is a dirty lie, and the problem appears to be related to hardlinking changing the ctime of the source file and confusing git into thinking that the files need to be re-checked-out. Bottom line, the fix is here:

git config --global core.trustctime false

Recent git branches

Sometimes I forget to delete my old git branches, and I forget which is what and who is how. And so on and so forth. So I do this!:

git reflog --date=local --all

This outputs a nifty list of all the recent commits and which branches they were too. w00t.

iOS timing

Again, do as I say not as I do. Here’s a quick thing I like to do:

NSTimeInterval start = CACurrentMediaTime();
NSTimeInterval end = CACurrentMediaTime();
NSLog(@"%f", end-start);

git format-patch and git am

Occasionally I need to send a git patch between computers. Yes, I know there are more efficient ways of doing this, but occasionally it needs to happen. So here’s how I do it:

git format-patch --binary master.. --stdout > the.patch

and how to apply it:

git am --signoff < the.patch

Sigh, GoDaddy

Well, it’s official. All my domains are now registered with GoDaddy. Why? Because while they have a website which makes my eyes bleed, they constantly have coupons like OMGWTF FREE DOMAIN TRANSFERS for 5 years. So, I pay nothing.

I’ll revisit who to transfer my domains to in about 5 years. Any suggestions until then?

Increase the max number of processes

I ran into this problem earlier today. This will increase the number of processes at a time, above the default 256. w00t.

echo 'limit maxproc 2000 2000' | sudo tee -a /etc/launchd.conf
sudo reboot