[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: line endings in text files and -kb

From: Paul Sander
Subject: Re: line endings in text files and -kb
Date: Thu, 11 Sep 2008 15:17:51 -0700

On Sep 11, 2008, at 1:45 PM, Ittay Dror wrote:

Paul Sander wrote:

On Sep 11, 2008, at 10:39 AM, Ittay Dror wrote:

Quite the opposite, YOU told cvs (via -kb) not to change the file, and it obeyed, i.e., you added the file with -kb and then committed it from MS, and on MS the file contained "/bin/ bash^M", and because of the -kb, cvs on Linux gave you EXACTLY the file committed on the MS machine which contains "/bin/bash^M"
what happened is this:
- there was no -kb
- a developer checks out the source tree and works with it, making some changes to .c files - not touching the file in question. the file is a startup script for a daemon - the build is ok on windows, so now he wants to check it is ok on linux
  - we don't want him to commit in order to test.
- so, he mounts the source tree from a linux machine and runs the build
 - this copies the file to the output of the build
- he wants to test the daemon, so he runs the startup script (the file in question) --> this fails. the file was checked out in windows, so now it has ^M

This sounds to me like it's not a CVS problem at all. The checkout to the Windows machine has the correct line endings for the platform. But then the sandbox is mounted from a different architecture using different line-ending conventions, the text files are interpreted as-is, and they break. It's unreasonable to assume that such an arrangement will work at all because the tools on the different platforms just don't interoperate well in this way (and rightly so).

The build on the Linux box should account for line endings if the source tree is mounted from Windows. Or the method used to copy the source tree from Windows to Linux could do the translation. Or you could check out on Linux and mount the Linux sandbox on Windows for testing, if Windows is more forgiving than Linux with regard to line endings. Or you could commit to a work branch and let CVS do the conversion while checking out into a second sandbox on Linux, and merge back to the parent branch after testing. Or you could reduce your aversion to committing unfinished code.
why does a tool like CVS take active action in changing a file for me and then i have to work around it?

CVS did the right thing for the client. The fact that the build worked on Windows proves this.

Your cross-mount from Linux broke the semantics of text files during your build on that platform. If the filesystem could tell the difference between text and binary files and performed the conversion instead, you wouldn't have this problem. But cross-mounts usually don't work that way.

if cvs had done nothing, well, ok. but it does the action that causes my chosen workflow to break, so now i have to take a counter action?

If you check the file out on the platform for which the text is ultimately destined, you would not have this problem. Check it out on Windows, it works on Windows. Check it out on Linux, it works on Linux. Under no circumstances, using CVS or not, should you expect text files on one platform to work on the other when simply mounted or copied bit-for-bit.

Throw in other platforms in which line-endings are CR (not CRLF as on Windows, or LF as on Unix), and the tools are *very* unforgiving, then you begin to see the complexity. So the tool gives text file semantics to text files stored on the client platform, which is a *much* simpler solution than storing the file in its original format and making every client (or worse, the native tool set) work around it.

checking a file as-is should have been the default, and changing its line endings should have been a nice feature to enable.

Go ahead and try it your way, and don't cry when you discover that you actually end up in a much worse condition. You'll discover that the problem is in your work flow, not with CVS.

This problem was solved multiple decades ago by assigning certain semantics to text files, then applying them universally in the various work flows where the bits cross platforms. Your choice to declare that text files should be bit-for-bit identical on all of your supported platforms violates this principle, and now you're paying the price. You'll be much happier if you design your work flow so that the cross-platform copies are made exclusively using tools that understand text file semantics.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]