abort: malformed response from CVS

[Markus Mayer]

I am getting an error similar to
http://bz.selenic.com/show_bug.cgi?id=4096. However, there are some
key differences:

* The repo I am trying to convert is simple, since it only contains a
single file (http://www.cybercom.net/~dcoffin/dcraw/RCS/dcraw.c,v)
with <500 commits
* This is on OS X 10.9
* The file system I am using to perform the conversion is case
sensitive, but the root file system is not.
* On a different OS X 10.9 machine, where ALL file systems are case
sensitive (including /), everything is working fine.
* I am using Mercurial 2.9.2.

Is there a way to diagnose where it's stumbling and whether the case
insensitive rootfs is indeed the culprit? Is it using temporary files
that live in /tmp or /var/tmp that could cause the problem? Is there a
way to tell it to put the temp files elsewhere?

I tried adding "print line" as suggested in bug 4096, but I don't
think the output is very helpful. However, I found some odd
differences comparing the "working" and "non-working" output.

Working
-------

$ hg convert . ../hg
initializing destination ../hg repository
connecting to /Users/markus/dcraw-holdingpen/cvs
scanning source...
collecting CVS rlog
482 log entries
creating changesets
100 Added the Kodak DCS Pro 14nx and SLR/c.
200 Color-balanced the Sigma SD9 based on the outdoor photo.\nAdded support f...
300 Halved memory usage by moving flip logic into write_ppm_tiff().
400 Support the Canon EOS 650D, Samsung NX20, and Sigma SD1.
468 changeset entries
sorting...
converting...
467 Initial revision
466 Use 2x2 squares instead of orthogonal neighbors.
465 Final 0.1 version.  Changed 2x2 blocks to weighted 3x3 neighborhood.
464 Changes for version 0.2:
463 Version 0.3
462 Version 0.4, just for testing
461 Version 0.8
460 Version 0.85
[...]
15 Fixed gravity sensor for Canon 5D Mark III with Firmware Version 1.1.x.
14 Support the Canon EOS 650D, Samsung NX20, and Sigma SD1.
13 Support the Samsung NX1000 and Sony RX100.
12 Measured saturation level on the Canon EOS-1D X.
11 Support the Canon EOS 6D, 1DX, SX50, Casio EX-ZR100, Fuji X-E1 & XF1,
10 Copied color matrices from DNG Converter 7.3.
9 Support the Fuji HS50, X20, X100S, SL1000, Samsung NX300,
8 Fix green spots when half-sizing Fuji X-Trans images.
7 initial checkin
6 Can now use "-M" with a DNG file that embeds a bad color matrix.
5 Support the Kodak 12-megapixel camera(s).
4 Use Little CMS version 2.
3 Added three temporary color matrices.
2 Fixed Sony A700 black level.
1 Support Lossy DNG files that lack an OpcodeList2 tag.
0 Removed redundant strings.

Failing
-------

$ hg convert . ../hg
initializing destination ../hg repository
connecting to /Volumes/InternalDisk/Development/dcraw-holdingpen/cvs
scanning source...
collecting CVS rlog
497 log entries
creating changesets
100 Added the Kodak DCS Pro 14nx and SLR/c.
200 Color-balanced the Sigma SD9 based on the outdoor photo.\nAdded support f...
300 Halved memory usage by moving flip logic into write_ppm_tiff().
400 Support the Canon EOS 650D, Samsung NX20, and Sigma SD1.
469 changeset entries
sorting...
converting...
468 Initial revision
467 Use 2x2 squares instead of orthogonal neighbors.
466 Final 0.1 version.  Changed 2x2 blocks to weighted 3x3 neighborhood.
465 Changes for version 0.2:
464 Version 0.3
463 Version 0.4, just for testing
462 Version 0.8
461 Version 0.85
[...]
15 Support the Canon EOS 650D, Samsung NX20, and Sigma SD1.
14 Support the Samsung NX1000 and Sony RX100.
13 Measured saturation level on the Canon EOS-1D X.
12 Support the Canon EOS 6D, 1DX, SX50, Casio EX-ZR100, Fuji X-E1 & XF1,
11 Copied color matrices from DNG Converter 7.3.
10 Support the Fuji HS50, X20, X100S, SL1000, Samsung NX300,
9 Fix green spots when half-sizing Fuji X-Trans images.
8 initial checkin
7 initial checkin
abort: malformed response from CVS

Somehow the changeset count is off by one (468 vs. 469). Also,
"initial checkin" appears twice (as commits 7 & 8) in the failing
case, but only once (commit 7) when it's working. Log entry count is
482 vs. 497.

On Ubuntu 14.04 (with Mercurial 2.8.2) the results are matching the
working OS X case (482 log entries, 468 changesets, "initial checkin"
as commit 7 only).

I can't think of anything that would be different between the two Macs
other than the rootfs being case-insensitive on the Mac where the
conversion is failing (the disk with the CVS and HG repos *IS*
case-sensitive, however). That being said, I am not sure if that
difference explains the behaviour observed.

Any tips how I could track this down further?

Thanks,
-Markus