Re: 5.2-stable regression for log axes still not fixed completely

[Ethan A M. <sf...@us...>]

On Tuesday, November 7, 2017 12:48:39 PM PST Achim Gratz wrote:
> Achim Gratz writes:
> > Right now it's somewhere in a 1800 pages PDF report, I'll have to
> > isolate the offending plots and get at the gnuplot code for that.  For
> > now I only know that those plots I've looked at didn't show any
> > unexpected defects, so the warning seems to be spurious as far as the
> > result goes.
> 
> This one _should_ be simple to fix: The warnings get issued when a
> previous plot used a reverse linear axis that is constrained to negative
> values and the next plot uses a logarithmic axis.  To reproduce:
> 
> --8<---------------cut here---------------start------------->8---
> set xrange [-2*pi:-pi] reverse
> set xrange [*:*] noreverse
> set log x
> --8<---------------cut here---------------end--------------->8---

Hold on a minute.
That first line is not a legal command in version 5, or at any rate
it is meaningless.  The "reverse" keyword only applies to autoscaling.
If that command is actually doing something, that's a bug all by itself.

... [thinking] ...

On the other hand, that spurious warning is generated even without
"reverse/noreverse" involved.  I think the actual error is that 
set xrange [*:*] turns on autoscaling but does not erase the previous
min/max.  Since they were both negative numbers, "set log" complains.

The code has a special check that handles the default axis range
setting [-10:10].  Maybe that check should be extended to cover all
cases of negative range limits?   Not sure.  But really I think the
problem is that "set auto" and "set range [*:*] don't actually clear
the previous limits.  Unfortunately I suspect that the "refresh" command
relies on that, so the best fix is not immediately clear.

	Ethan



> 
> The error does not happen if the max of the first xrange is positive.
> It seems that switching on autoscale does not correctly invalidate some
> of the internal variables set up by the first xrange.
> 
> The explanation for why this error is not having consequences on my
> plots is that the correct range for the (now logarithmic) axis will
> always be set later on.
> 
> 
> Regards
> Achim.