The first repository branch, "Windows linestyle fix."

[Daniel J S. <dan...@ie...>]

On 10/25/2017 02:48 PM, Daniel J Sebald wrote:
> On 10/25/2017 12:16 PM, Eric S. Raymond wrote:
[snip]
>> I'm going to to do some more analysis on the 3.7 branch, but the odds
>> that we'll have to write it off look pretty high.
> 
> I'll look at this a bit more.  For those with historical knowledge, I'm 
> going to write some checkin comments here that come from searching, but 
> only those that seem like they could give a clue about when the first 
> branch was done.  (I'm using qgit, which has some nice searching and 
> file features.)  Let me know if they ring a bell:
> 
> "branch"
> New function gp_strdup, from cvs MAIN branch.
> Updated from cvs MAIN branch
> New files from my branched-off 'monster patch'
> Merged the pm3d branch back to MAIN
> Merge axis-branch into MAIN. Lots of changes all over the places
> a few changes after merging the axis branch
> make the branched version compile under OS/2
> Updated from aquaterm cvs bugfix branch.
> 
> "merge"
> Final fixes for merger.
> Update from current automake (lost in merge).
> Some cleanups/fixes after the merge.

Here's a bit of a clue about the branch point for that first branch.  I 
noted that the two files that had changed with the "Windows linestyle 
fix." checkin are:

=============================================================================
RCS file: /cvsroot/gnuplot/gnuplot/win/Attic/wgnuplib.h,v
Working file: win/wgnuplib.h
head: 1.6
branch:
----------------------------
revision 1.4.2.1
date: 1999/08/19 14:13:09;  author: lhecking;  state: Exp;  lines: +4 -1
Windows linestyle fix.
=============================================================================

=============================================================================
RCS file: /cvsroot/gnuplot/gnuplot/win/Attic/wgraph.c,v
Working file: win/wgraph.c
head: 1.7
branch:
----------------------------
revision 1.5.2.1
date: 1999/08/19 14:13:09;  author: lhecking;  state: Exp;  lines: +12 -1
Windows linestyle fix.
=============================================================================

However, compare the date with the following major modification which 
moved those files from the win/ directory to the win/src directory (and 
remember in CVS the "move" is actually as though these are completely 
different files, with new instantiation):

=============================================================================
RCS file: /cvsroot/gnuplot/gnuplot/win/Attic/wgnuplib.h,v
Working file: win/wgnuplib.h
head: 1.6
branch:
----------------------------
revision 1.6
date: 1999/03/26 22:09:29;  author: lhecking;  state: dead;  lines: +1 -1
Moved to src/win/.
----------------------------
=============================================================================

So, that modification "Windows linestyle fix" is dated five months 
*after* the file was moved, effectively terminated.  There is no way the 
branch point could be near 1999/08/19 where I was searching.  So, no 
surprise that search failed.

I've searched back much earlier.  I'm not familiar with CVS versioning, 
does the "revision 1.4.2.1 date: 1999/08/19 14:13:09;" mean that this 
particular revision was derived from "revision 1.4"?  I.e., the user 
checked out "revision 1.4" as a start for creating the branch?  In this 
case revision 1.4 was of

----------------------------
revision 1.4
date: 1998/12/10 18:30:13;  author: lhecking;  state: Exp;  lines: +0 -2
branches:  1.4.2;
Remove two defines.
----------------------------

So, maybe that is where I should look, around 1998/12/10.  (I should 
probably just write a script to compare against all versions.) ...
Nope.  If I'm understanding correctly, based upon the "branches" number 
of 1.4.2.1, "Windows linestyle fix" could have happened any time between 
revision 1.5 (1999/02/14 18:13:55) and revision 1.4 (1998/12/10 
18:30:13).  The logical branch points in that range would be

date: 1999/01/12 14:03:53;  author: lhecking;  state: Exp;  lines: +4 -5
Version 3.7, patchlevel 0, update date string.

and

date: 1999/01/12 13:58:02;  author: lhecking;  state: Exp;  lines: +3 -3
Updated to 3.7.

but again, those don't seem to match well.

Well, here's one encouraging result:  I've done the directory comparison 
between CVS versions.  If I pick CVS directory to be

cvs update -rGNUPLOT_RELEASE_3_7_0

and compare against the git "Windows linestyle fix." directory it 
creates pretty good agreement but it looks like all the RCSid are rather 
different.  Just to illustrate, here a diff of one of the files that 
supposedly changes both within CVS and across CVS and git:

sebald@ ~/gnuplot/git_translation/branch_point_search/gnuplot $ cvs diff 
-u -r1.4 -r1.4.2.1 win/wgnuplib.h
Index: win/wgnuplib.h
===================================================================
RCS file: /cvsroot/gnuplot/gnuplot/win/Attic/wgnuplib.h,v
retrieving revision 1.4
retrieving revision 1.4.2.1
diff -u -r1.4 -r1.4.2.1
--- win/wgnuplib.h	10 Dec 1998 18:30:13 -0000	1.4
+++ win/wgnuplib.h	19 Aug 1999 14:13:09 -0000	1.4.2.1
@@ -1,5 +1,5 @@
  /*
- * $Id: wgnuplib.h,v 1.4 1998/12/10 18:30:13 lhecking Exp $
+ * $Id: wgnuplib.h,v 1.4.2.1 1999/08/19 14:13:09 lhecking Exp $
   */

  /* GNUPLOT - win/wgnuplib.h */
@@ -320,6 +320,9 @@
  	HPEN	hbpen;					/* border pen */
  	HPEN	hapen;					/* axis pen */
  	HPEN	hpen[WGNUMPENS];		/* pens */
+#if 1 /* HBB 980118: new try ... */
+        HPEN    hsolidpen[WGNUMPENS];           /* solid pens (for 
point symbols) */
+#endif
  	LOGPEN	colorpen[WGNUMPENS+2];	/* logical color pens */
  	LOGPEN	monopen[WGNUMPENS+2];	/* logical mono pens */
  	COLORREF background;			/* background color */

sebald@ ~/gnuplot/git_translation/branch_point_search $ diff -ur 
--exclude=".git" gnuplot/win/wgnuplib.h branch_reference/win/wgnuplib.h
--- gnuplot/win/wgnuplib.h	1998-12-10 12:30:13.000000000 -0600
+++ branch_reference/win/wgnuplib.h	2017-10-25 16:05:19.318473813 -0500
@@ -320,6 +320,9 @@
  	HPEN	hbpen;					/* border pen */
  	HPEN	hapen;					/* axis pen */
  	HPEN	hpen[WGNUMPENS];		/* pens */
+#if 1 /* HBB 980118: new try ... */
+        HPEN    hsolidpen[WGNUMPENS];           /* solid pens (for 
point symbols) */
+#endif
  	LOGPEN	colorpen[WGNUMPENS+2];	/* logical color pens */
  	LOGPEN	monopen[WGNUMPENS+2];	/* logical mono pens */
  	COLORREF background;			/* background color */

The discouraging part, though, is that when I compare CVS 
GNUPLOT_RELEASE_3_7_0 against git master commit "Updated to 3.7.", it's 
not a good match at all.  Shouldn't it be a good match?

I think the key here is to first figure out and reconstruct what would 
be tag 3.7.0 in the git repository, so that it coincides with 
GNUPLOT_RELEASE_3_7_0.  It might be the something is corrupted in the 
master branch in that first few years of checkins.  My guess would be 
that release 3.7.0 was on the trunk, then "Windows linestyle fix" was a 
branch, which produced 3.7.1, 3.7.2, etc.  I can see why this is so 
difficult to reconstruct now.  Tags are the only sort of reliable CVS 
construct, so I suppose we should get those important ones working first.

Dan