cvs2svn: Got Malformed RCS delta error while reproducing newer revison file from older revision file.

[futatuki]

(This was reported by @jun66j5 on twitter, in Japanese language, with larger example.)

It seems cvs2svn cannot handle some RCS files which contains "a" command with in complete (not end with '\n') line before the last line in the target file.

Here is an example file irregular.txt,v:

head    1.5;
access;
symbols;
locks; strict;
comment @# @;


1.5
date    2021.11.26.04.16.45;    author futatuki;        state Exp;
branches;
next    1.4;

1.4
date    2021.11.26.04.14.23;    author futatuki;        state Exp;
branches;
next    1.3;

1.3
date    2021.11.26.03.46.06;    author futatuki;        state Exp;
branches;
next    1.2;

1.2
date    2021.11.26.03.41.39;    author futatuki;        state Exp;
branches;
next    1.1;

1.1
date    2021.11.26.03.34.28;    author futatuki;        state Exp;
branches;
next    ;


desc
@create a new file irregular.txt
@


1.5
log
@r1.5:
@
text
@aaa
bbb
ccc
hhh@


1.4
log
@r1.4:
@
text
@d4 1
a4 1
ggg@


1.3
log
@r1.3: s/eee/fff/
@
text
@d3 1
a3 1
fff@


1.2
log
@r1.2: s/ddd/eee/
@
text
@d3 1
a3 1
eee@

1.1
log
@r1.1
@
text
@d3 1
a3 1
ddd@

For each revision diff text, the last 'a' command adds incomplete line. Especially in r1.3 from r1.4, this diff text replaces an existing comlete line into an incomplete line, and as a result, this line connect the next existing line (although I don't believe CVS/RCS can produce such a curious diff, but there exists other than this in a real use case, and CVS/RCS can handle it).

With a sample CVS repository contains this irregular.txt,v only (except RCSROOT), cvs2svn produces an error:

$  cvs2svn -s irregular-rcs-test-broken.svn /home/cvs/irregular-rcs-test
Writing temporary files to '/tmp/cvs2svn-Tlxzju'
Be sure to use --tmpdir='/tmp/cvs2svn-Tlxzju' if you need to resume this conversion.
----- pass 1 (CollectRevsPass) -----
Examining all CVS ',v' files...
/home/cvs/irregular-rcs-test/irregular.txt,v
/home/cvs/irregular-rcs-test/CVSROOT/checkoutlist,v

  ...

Starting Subversion r7 / 7
ERROR: Malformed RCS delta in /home/cvs/irregular-rcs-test/irregular.txt,v, revision 1.5: Deletion beyond file end

With the incomplete Subversion repository irregular-rcs-test-broken.svn, r1.1 to r1.4 revision files of irregular.txt exactly equal to ones of original CVS repo's.

Although I also think those RCS delta was malformed, I think it is better to continue to build new revisions for Subversion repo, because newer revision files before mailformed delta are correct files.

I'll submit a PR for the issue.

[mhagger]

Thanks for the careful report and the research. Your conclusions could be corroborated if you try converting the same file, first with the --use-rcs and then with the --use-cvs option, to see if those conversions give the result that you expect. If they do, then that's pretty conclusive proof that the --use-internal-co version needs to be adjusted. If they fail, or if they give different results, then the analysis might need to be adjusted.

[futatuki]

Thank you for the suggestion. Both of with --use-rcs option and with --use-cvs option, cvs2svn produced as I expected, all r1.1 to r1.5 rivision of irregular.txt file checkout correctly from converted repository as r3 to r7.

Diff of dump files with CVS and with RCS are just UUID and svn:date of r1.

--- irregular-rcs-test-cvs-dump.txt     2021-12-02 19:59:31.895085000 +0900
+++ irregular-rcs-test-rcs-dump.txt     2021-12-02 19:59:46.421823000 +0900
@@ -1,6 +1,6 @@
 SVN-fs-dump-format-version: 2
 
-UUID: 7caf5ac0-5e53-ec11-9708-001c4272a9f0
+UUID: 78e207b6-5e53-ec11-9708-001c4272a9f0
 
 Revision-number: 0
 Prop-content-length: 56
@@ -9,7 +9,7 @@
 K 8
 svn:date
 V 27
-2021-12-02T10:58:16.104635Z
+2021-12-02T10:57:58.781240Z
 PROPS-END
 
 Revision-number: 1

And with --use-internal-co,

With the incomplete Subversion repository irregular-rcs-test-broken.svn, r1.1 to r1.4 revision files of irregular.txt exactly equal to ones of original CVS repo's.

This was incorrect. Only r1.1 == r3 was the same. All other revisions are broken.

After applying PR #19, cvs2svn with --use-internal-co successfully convert the repository, and its dump file differ only repos UUID and svn:date of r1.