[Gnu-arch-users] File history and line annotation

[Mikhael Goikhman]

On 06 Feb 2005 12:29:05 -0500, Stefan Monnier wrote:
> 
> It may be.  I also never care to "blame".  Actually I don't even want to
> "annotate" either.  I don't really want the output that "cvs annotate"
> gives me, what I want instead is to trace the live of a piece of code.

Yes, pure "annotate" is of limited usefulness. Still, "tla annotate" may
be done better than "cvs annotate", it may show a summary of line groups.

> I.e. from a set of code lines, I want to see all the diff+log that resulted
> in those lines.  Also, often enough the diff+log is not what I want because
> it just corresponds to a re-indent or a move, so I often also want to see
> the parents of those diffs (i.e. the diff+log that resulted in the text that
> was then changed by those diffs), their parents's parents, ...
> until I really find the diff+log where the code was originally added.

It seems that you speak about the complete file history. Like this:

  http://arch.debian. 
org/cgi-bin/archzoom.cgi/address@hidden/zsh--cscvs--0--patch-3645/README?template=plain?history

or this (with renames and version crossing):

  http://arch.debian. 
org/cgi-bin/archzoom.cgi/address@hidden/postgresql--no-cluster-ports-single--8.0--patch-10/postgresql8.0-8.0.0/debian/control?history

I don't think it's possible to present such information clearly including
hyperlinks in the command line interface. The axp tool however has such
"history" command that may help to analyze a file.

[Baz is likely to have such command too. However, they want to use a
confusing name for it, "log", that would conflict with the term used in
arch for revision log, as in `tla help | grep log` commands. I strongly
oppose to this conflicting name, in favor of a more intuitive "history".]

> With CVS I typically do this manually: `cvs annotate' to find the relevant
> revision, then `cvs diff' to get the corresponding diff and `cvs log' to see
> the corresponding message.  Then if the diff is just moving the thing
> around, I go back to `cvs annotate' the preceding revision, ...
> It's all very inefficient for me.

It's quite possible you may be more efficient with a web based interface.

Or probably a GUI. Try archway 0.1.1, it should give what you need too,
a history of any selected file or directory. I.e. conveniently bunched
revision descriptions with operations for every revision that touched the
file: to visually show changeset or log, or to get changeset or tree.

> In Arch it'll be even worse because the data structure we have is not a
> good match to "tla annotate".  I really wish someone will write
> "tla trace-code" instead of writing yet-another "tla annotate".

Can you formalize such "trace-code" command, its input and output?

I think what you really describe is a web based or a GUI based tool with
a fast navigation in the file history, from one diff to another.

Or maybe even an "annotate"-like look of the file content with an ability
to remove layers (i.e. diffs) from the file. This functionality will be
available in archway together with "annotate" support.

Regards,
Mikhael.