Execute a temp copy of update_depot_tools.bat

Execute a temp copy of update_depot_tools.bat

Windows will sometimes freak out if a file is rewritten while
it's being executed.  That can happen when update_depot_tools.bat
runs.

R=iannucci@chromium.org,maruel@chromium.org
BUG=

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=189429

[szager1, 2013-03-20 20:57:54]

Pulling out this functionality from here:

https://codereview.chromium.org/12865010/

[iannucci, 2013-03-20 21:06:14]

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat
File update_depot_tools.bat (right):

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode10
update_depot_tools.bat:10: :: Windows sometimes freaks out if a file is
overwritten while it's being
'sometimes'?

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode14
update_depot_tools.bat:14: COPY /Y %~dp0update_depot_tools.bat
%~dp0update_depot_tools_tmp.bat >nul
is %~dp guaranteed to end with a r'\' ?

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode16
update_depot_tools.bat:16: %~dp0update_depot_tools_tmp.bat %*
Should we goto :EOF after this? I forget how batch subinvocations work...

[M-A Ruel, 2013-03-20 21:06:37]

https://codereview.chromium.org/12755033/diff/1/.gitignore
File .gitignore (right):

https://codereview.chromium.org/12755033/diff/1/.gitignore#newcode22
.gitignore:22: # Ignore temporary locations for auto-updating scripts.  These
are necessary
Not needed.

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat
File update_depot_tools.bat (right):

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode14
update_depot_tools.bat:14: COPY /Y %~dp0update_depot_tools.bat
%~dp0update_depot_tools_tmp.bat >nul
Please use %TEMP%\update_depot_tools_%RANDOM%.bat, so the likelihood of 2
gclient process breaking each other lowers from 100% to 1/~10000, since we can't
trust the random number generator to be efficient. You could use
%RANDOM%_%RANDOM% if you prefer but since it's a pseudo-random generator, it
won't help much.

[szager1, 2013-03-20 21:19:10]

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat
File update_depot_tools.bat (right):

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode10
update_depot_tools.bat:10: :: Windows sometimes freaks out if a file is
overwritten while it's being
On 2013/03/20 21:06:14, iannucci wrote:
> 'sometimes'?

I haven't explored every nuance of this behavior (e.g., what happens with
subprocesses, or exec from bash, or ...).  But I can reproduce the bad
situation, if necessary.

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode14
update_depot_tools.bat:14: COPY /Y %~dp0update_depot_tools.bat
%~dp0update_depot_tools_tmp.bat >nul
On 2013/03/20 21:06:37, Marc-Antoine Ruel wrote:
> Please use %TEMP%\update_depot_tools_%RANDOM%.bat, so the likelihood of 2
> gclient process breaking each other lowers from 100% to 1/~10000, since we
can't
> trust the random number generator to be efficient. You could use
> %RANDOM%_%RANDOM% if you prefer but since it's a pseudo-random generator, it
> won't help much.

That means that I have to fix every occurence of %~dp0 in this file.  Which I
can do, but I was hoping to leave a smaller footprint.

Do we really claim to support multiple concurrent invocations of gclient?

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode16
update_depot_tools.bat:16: %~dp0update_depot_tools_tmp.bat %*
On 2013/03/20 21:06:14, iannucci wrote:
> Should we goto :EOF after this? I forget how batch subinvocations work...

Empirical evidence suggests that it's not necessary, but I'll add it anyways,
just to be safe.

[iannucci, 2013-03-20 21:33:14]

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat
File update_depot_tools.bat (right):

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode10
update_depot_tools.bat:10: :: Windows sometimes freaks out if a file is
overwritten while it's being
On 2013/03/20 21:19:10, szager1 wrote:
> On 2013/03/20 21:06:14, iannucci wrote:
> > 'sometimes'?
> 
> I haven't explored every nuance of this behavior (e.g., what happens with
> subprocesses, or exec from bash, or ...).  But I can reproduce the bad
> situation, if necessary.

I was meaning that the 'sometimes' can just be omitted :). I believe that batch
scripts are modifiable 'live', so if svn pulls a change which moves all the
lines in the batch file, cmd.exe could end up skipping a line or executing a
line (or lines(s)) multiple times. See 'Compound Modification' in
http://en.wikipedia.org/wiki/Self-modifying_code

[M-A Ruel, 2013-03-20 21:51:22]

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat
File update_depot_tools.bat (right):

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode10
update_depot_tools.bat:10: :: Windows sometimes freaks out if a file is
overwritten while it's being
On 2013/03/20 21:33:14, iannucci wrote:
> I believe that batch
> scripts are modifiable 'live' (...)

Exact, cmd.exe keeps a byte offset and re-reads the file on every line. No
wonder batch files are slow as molasse.

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode14
update_depot_tools.bat:14: COPY /Y %~dp0update_depot_tools.bat
%~dp0update_depot_tools_tmp.bat >nul
On 2013/03/20 21:19:10, szager1 wrote:
> On 2013/03/20 21:06:37, Marc-Antoine Ruel wrote:
> > Please use %TEMP%\update_depot_tools_%RANDOM%.bat, so the likelihood of 2
> > gclient process breaking each other lowers from 100% to 1/~10000, since we
> can't
> > trust the random number generator to be efficient. You could use
> > %RANDOM%_%RANDOM% if you prefer but since it's a pseudo-random generator, it
> > won't help much.
> 
> That means that I have to fix every occurence of %~dp0 in this file.  Which I
> can do, but I was hoping to leave a smaller footprint.
> 
> Do we really claim to support multiple concurrent invocations of gclient?

I don't mind about not using random but you should still put it in TEMP.

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode16
update_depot_tools.bat:16: %~dp0update_depot_tools_tmp.bat %*
On 2013/03/20 21:19:10, szager1 wrote:
> On 2013/03/20 21:06:14, iannucci wrote:
> > Should we goto :EOF after this? I forget how batch subinvocations work...
> 
> Empirical evidence suggests that it's not necessary, but I'll add it anyways,
> just to be safe.

No it's not necessary. As long as there is not 'call' statement, the control is
deferred to the called batch file.

[szager1, 2013-03-20 21:53:52]

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat
File update_depot_tools.bat (right):

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode10
update_depot_tools.bat:10: :: Windows sometimes freaks out if a file is
overwritten while it's being
On 2013/03/20 21:33:14, iannucci wrote:
> On 2013/03/20 21:19:10, szager1 wrote:
> > On 2013/03/20 21:06:14, iannucci wrote:
> > > 'sometimes'?
> > 
> > I haven't explored every nuance of this behavior (e.g., what happens with
> > subprocesses, or exec from bash, or ...).  But I can reproduce the bad
> > situation, if necessary.
> 
> I was meaning that the 'sometimes' can just be omitted :). I believe that
batch
> scripts are modifiable 'live', so if svn pulls a change which moves all the
> lines in the batch file, cmd.exe could end up skipping a line or executing a
> line (or lines(s)) multiple times. See 'Compound Modification' in
> http://en.wikipedia.org/wiki/Self-modifying_code

OK, I removed 'sometimes'.  Why give false hope?

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode14
update_depot_tools.bat:14: COPY /Y %~dp0update_depot_tools.bat
%~dp0update_depot_tools_tmp.bat >nul
On 2013/03/20 21:06:14, iannucci wrote:
> is %~dp guaranteed to end with a r'\' ?

Hmmm... I've never seen it any other way.

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode14
update_depot_tools.bat:14: COPY /Y %~dp0update_depot_tools.bat
%~dp0update_depot_tools_tmp.bat >nul
On 2013/03/20 21:06:14, iannucci wrote:
> is %~dp guaranteed to end with a r'\' ?

Hmmm... I've never seen it any other way.  Theoretically, if the script was at
the top directory level, I suppose that might not be true.  But if that happens,
I'd just as soon this script died, since you're clearly in a bad state.

[szager1, 2013-03-20 22:03:44]

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat
File update_depot_tools.bat (right):

https://codereview.chromium.org/12755033/diff/1/update_depot_tools.bat#newcode14
update_depot_tools.bat:14: COPY /Y %~dp0update_depot_tools.bat
%~dp0update_depot_tools_tmp.bat >nul
On 2013/03/20 21:51:22, Marc-Antoine Ruel wrote:
> On 2013/03/20 21:19:10, szager1 wrote:
> > On 2013/03/20 21:06:37, Marc-Antoine Ruel wrote:
> > > Please use %TEMP%\update_depot_tools_%RANDOM%.bat, so the likelihood of 2
> > > gclient process breaking each other lowers from 100% to 1/~10000, since we
> > can't
> > > trust the random number generator to be efficient. You could use
> > > %RANDOM%_%RANDOM% if you prefer but since it's a pseudo-random generator,
it
> > > won't help much.
> > 
> > That means that I have to fix every occurence of %~dp0 in this file.  Which
I
> > can do, but I was hoping to leave a smaller footprint.
> > 
> > Do we really claim to support multiple concurrent invocations of gclient?
> 
> I don't mind about not using random but you should still put it in TEMP.

Done.

[M-A Ruel, 2013-03-20 22:05:35]

lgtm

[commit-bot: I haz the power, 2013-03-20 22:08:39]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/szager@chromium.org/12755033/15001

[commit-bot: I haz the power, 2013-03-20 22:08:43]

Failed to apply patch for depot_tools/update_depot_tools.bat:
While running patch -p1 --forward --force --no-backup-if-mismatch;
  patching file depot_tools/update_depot_tools.bat
  Hunk #1 FAILED at 7.
  Hunk #2 FAILED at 22.
  Hunk #3 FAILED at 51.
  3 out of 3 hunks FAILED -- saving rejects to file
depot_tools/update_depot_tools.bat.rej

Patch:       depot_tools/update_depot_tools.bat
Index: update_depot_tools.bat
diff --git depot_tools/update_depot_tools.bat depot_tools/update_depot_tools.bat
index
f880d9b895a384c616915c6d1f4ce5bc7c063319..20ac0d7cd45bf214220a1eca40708e7f4cc9358a
100644
--- a/depot_tools/update_depot_tools.bat
+++ b/depot_tools/update_depot_tools.bat
@@ -7,13 +7,25 @@
 
 setlocal
 
+:: Windows freaks out if a file is overwritten while it's being executed.  Copy
+:: this script off to a temporary location and reinvoke from there before
+:: running any svn or git commands.
+IF %~nx0==update_depot_tools.bat (
+  COPY /Y %~dp0update_depot_tools.bat %TEMP%\update_depot_tools_tmp.bat >nul
+  if errorlevel 1 goto :EOF
+  %TEMP%\update_depot_tools_tmp.bat %~dp0 %*
+)
+
+set DEPOT_TOOLS_DIR=%1
+SHIFT
+
 set GIT_URL=https://chromium.googlesource.com/chromium/tools/depot_tools.git
 
 :: Will download svn and python.
 :: If you don't want to install the depot_tools version of these tools, remove
 :: the 'force' option on the next command. The tools will be installed only if
 :: not already in the PATH environment variable.
-call "%~dp0bootstrap\win\win_tools.bat" force
+call "%DEPOT_TOOLS_DIR%bootstrap\win\win_tools.bat" force
 if errorlevel 1 goto :EOF
 :: Now clear errorlevel so it can be set by other programs later.
 set errorlevel=
@@ -22,19 +34,19 @@ set errorlevel=
 IF "%DEPOT_TOOLS_UPDATE%" == "0" GOTO :EOF
 
 :: We need either .\.svn\. or .\.git\. to be able to sync.
-IF EXIST "%~dp0.svn\." GOTO :SVN_UPDATE
-IF EXIST "%~dp0.git\." GOTO :GIT_UPDATE
+IF EXIST "%DEPOT_TOOLS_DIR%.svn\." GOTO :SVN_UPDATE
+IF EXIST "%DEPOT_TOOLS_DIR%.git\." GOTO :GIT_UPDATE
 echo Error updating depot_tools, no revision tool found.
 goto :EOF
 
 
 :SVN_UPDATE
-call svn up -q "%~dp0."
+call svn up -q "%DEPOT_TOOLS_DIR%."
 goto :EOF
 
 
 :GIT_UPDATE
-cd /d "%~dp0."
+cd /d "%DEPOT_TOOLS_DIR%."
 call git config remote.origin.fetch > NUL
 if errorlevel 1 goto :GIT_SVN_UPDATE
 for /F %%x in ('git config --get remote.origin.url') DO (
@@ -51,6 +63,6 @@ call git rebase -q origin/master > NUL
 goto :EOF
 
 :GIT_SVN_UPDATE
-cd /d "%~dp0."
+cd /d "%DEPOT_TOOLS_DIR%."
 call git svn rebase -q -q
 goto :EOF

[szager1, 2013-03-20 22:13:26]

Committed patchset #4 manually as r189429.

[M-A Ruel, 2013-03-20 22:17:30]

On 2013/03/20 22:13:26, szager1 wrote:
> Committed patchset #4 manually as r189429.

Did you commit from a windows machine to be sure it still works?

:D

[szager, 2013-03-20 22:22:52]

On Wed, Mar 20, 2013 at 3:17 PM, <maruel@chromium.org> wrote:

> On 2013/03/20 22:13:26, szager1 wrote:
>
>> Committed patchset #4 manually as r189429.
>>
>
> Did you commit from a windows machine to be sure it still works?
>
> :D
>

Yechh.  No, I didn't.  But I just updated my win checkout, and it works for
me.