Re: wrt 'Make doc crashing on my PC'

[Keith OHara]

Phil Holmes <mail <at> philholmes.net> writes:

> ----- Original Message ----- 
> > that reason at least on my machine. It seems to me therefore that 
2.17.18
> > do not expose the issue but 2.17.19 does.
> 
> That stacks up well with my earlier investigations.  Git besect 
identified 
> 35d5f2e2ff40e0bd35cf00f22c2428eac354e566 as the first bad commit, and 
that 
> commit occurred between 17.18 and 17.19.  

> However, I can only get a crash when I run make doc, so wasn't 100% sure 
> this was really the cause. 

My revert of my first attempt at better tempo-mark spacing (issue 3279)
was also between 17.18 and 17.19, and this first attempt of mine avoided
issue 3409 for a couple minor versions.

When I run LilyPond on orchestra.ly using the debugger, the first time I
see a stack layout matching Petr's backtrace is during the horizontal
spacing step, when LilyPond is estimating the vertical position of the
"Presto" tempo mark, in order to see if it is necessary to allow 
horizontal space for it.  That tempo mark contains all 25 StaffSymbols
in the score as its as side-suport-elements, the objects whose positions
it uses to determine its own position.

This means LilyPond tries to do a tentative vertical spacing of the 
staves, which depends in turn on the vertical position of the "Presto"
mark through various paths.  With debugging on, I get four messages 
"programming error: cyclic dependency: calculation-in-progress 
 encountered for #'adjacent-pure-heights (VerticalAxisGroup)"
when executing side-position-interface.cc:239, immediately before 
the call to the Skyline_pair constructor on line 240 that appears
in Petr's stack trace.

These error messages (issue 3409) have been around for a few major 
versions.  My first attempt at issue 3279 skipped the step of 
estimating the vertical positions of tempo marks, avoiding the error
messages, and possibly avoiding the crash.