On Sun, Oct 06, 2013 at 12:09:50PM -0400, Richard Heck wrote:
>
> I just wanted to call everyone's attention to this bug:
> http://www.lyx.org/trac/ticket/8854
> It appears to be a very nasty dataloss bug in 2.1-beta1. What seems
> to be happening is that, on save, the LyX file is corrupted in very
> weird ways, with random parts of the file being replaced by things
> like "mathnorma" or "Plain La"---fragments of things that do make
> sense in other contexts.
>
> A report of what looks like the same bug in 2.0.6 is here:
> http://tex.stackexchange.com/questions/134009/lyx-inserts-and-removes-words-randomly
> That was on Windows, as well, whereas #8854 is on Ubuntu. Another
> difference is that #8854 seems to show up only on export. The
> original report was of corruption in the exported 2.0 file, but by
> looking in the temporary directory we found that the 2.1 file was
> already corrupt. That makes it sound as if the problem has to do
> with background export. But the 2.0.6 report seems to concern just
> saving the file and not export.
>
> The reporter has done some excellent work to help us figure this
> out, but neither Pavel nor I have been able to reproduce---though
> the reporter sees it all the time. It would really, really help if
> someone could manage to reproduce this.
>
> Obviously, it would also help if someone has a good idea what might
> cause this.
I can reproduce this bug on several platforms and I am surprised that
nobody else can.
The problem here is that to_utf8() (and most probably all other
conversion routines) is not thread safe, maybe due to the static
output buffer in iconv_convert().
This is demonstrated by the attached quick and dirty patch. Without
this patch, I get a good export only about once every ten trails.
With the patch the exported file is always correct.
--
Enrico
diff --git a/src/support/docstring.cpp b/src/support/docstring.cpp
index 727e1f5..db4f513 100644
--- a/src/support/docstring.cpp
+++ b/src/support/docstring.cpp
@@ -107,7 +107,7 @@ docstring const from_utf8(string const & utf8)
string const to_utf8(docstring const & ucs4)
{
- vector<char> const utf8 = ucs4_to_utf8(ucs4.data(), ucs4.size());
+ QByteArray utf8 = toqstr(ucs4).toUtf8();
return string(utf8.begin(), utf8.end());
}
