Merge branch 'master' into alt

These are two identical ways to get the same result.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: add a commented-out call to synch everything

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: fix handling of empty lines from findcurve

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Revert "simplex wip: compiles"

This reverts commit 657118ffcf7627d9c97d66065fba358248bbf860.

Revert "curveopt: some debug etc."

This reverts commit 35a742a8217c9b54eeea81a8e5c139295f528de8.

Revert "curveopt: try simple, but no good"

This reverts commit 7b5370bb5115b5179c4fb36093677330ef909e59.

curveopt: try simple, but no good

It seems to just shrink the simplex and then exit.  I think it doesn't
manage to find the narrow trench in configuration space in which
possible improvements lie.

Also tried gsl_multimin_fminimizer_nmsimplex but this is even slower
than siman.  So we will go back to siman now.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: some debug etc.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: add a commented-out call to synch everything

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: fix handling of empty lines from findcurve

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

simplex wip: compiles

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Revert "simplex wip: use gsl_vector_get for X, for abandonment"

This reverts commit 29cb20f6187aeed6da1f1a7efa3473ee46015ece.

simplex wip: use gsl_vector_get for X, for abandonment

This is too painful because it breaks POINT, which everyone expects to
be able to inspect with []

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: show it in a better way

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: use new thing

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: bespoke debugging

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: parallelise findcurve processes

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: use a fresh findcurve process for each curve

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: some more checks

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: better costs, seems more plausible now

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: ensure |EF| and |HG| are positive

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: new approach

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: wip, demos more of the same bug

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: wip, observe a bug

This algorithm encourages a segment with a curve either side of it to
be shorter than its neighbours.  In this test case there is another
bug: the cp[1] (F) and cp[2] (G) vectors are wrong, exaggerating the
effect of the bug.

To fix the wrong encouragement bug, I think we will have to split the
penalty for curvature from the penalty for density variation.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: wip

Makefile: fix spelling of COPTIMISE

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: test a bit

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visual: reduce nv and nw for debugging

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt.py: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt.py: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

moebius: break out moebez.py

curveopt is going to want these.  We need to move it into a separate
file to avoid a circular import dependency.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: rename from helixish

We're going to want to reuse the findcurve invocation machinery.  Most
of the rest of this is specific to the abandoned helixish approach, so
is due to be hacked away.  For now just rename this unused file.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

moebius: break out DiscreteBezier (nfc)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

moebius: pass nt and it to MoebiusHalf, rather than just t

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: symbolic: compiles

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: symbolic: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: symbolic: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: symbolic: wip, before abandon provision for fdf

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: symbolic: wip, before go back to conditional in C

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

curveopt: symbolic: wip

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Revert "RecursiveBezierishCurve: Try it"

This reverts commit 9a77b7a72b35200447e25c105ae383a1186a3899.

RecursiveBezierishCurve: Try it

This does not seem to make any difference.  How interesting.  I
suspect that in fact this construction arranges, somehow, to get the
same cubics for the two halves as for the original.  But I CBA to do
the algebra to check.

Ah well, we will revert this and try something else.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: Abandon

Remove calls to Helixish and hence to findcurve et al.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Revert "helixish: try three paths"

This reverts commit 7de6976b27662081219291a52947d19bee422d44.

helixish: try three paths

This shows that this approach does not work.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Revert "visual: debugging with one in darker blue"

This reverts commit 756c4104abae08a963de109dd9a2ea4659698f40.

Revert "helixish: attempt at the whole thing"

This reverts commit 906edec3ef60cf3e0567c713b2a68688edb634d4.

helixish: attempt at the whole thing

This does not really work.  Should revert and try bits of it
one at a time, etc.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: seems to be able to solve

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: seems to be able to solve but answer is wrong due to wrong energy

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging changes

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: fix .iters_fixed_T

If this is zero, gsl_siman_solve does not do anything.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging changes

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging changes

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visual: debugging changes

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: fix sign problems

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: remember to transform back from findcurve basis (!)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: try_s and try_t now always >= 0

Since we add tau if necessary

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visual: debugging changes

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visual: debugging with one in darker blue

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: ensure s and t are >0

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: dist_pq_plane should be the distance in the dp/dq plane

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: convert r from polynomial back to our unscaled plane coordinates

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: add a trivial comma (nfc)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: rename vdbg_railway_angle for clarity (nfc)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visdebug: make circle of variable size

Existing arrows have their circle sizes adjusted according to the
lengths of the arrows.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visdebug: introduce vdbg().circle() (nfc)

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: get sense of phi right

In "Railway problem", phi is the direction we _leave_ Q.  Whereas in
our original representation, we have a direction vector in the
positive sense of the parameter (ie, for Q, the arrival direction).

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: angles are unconventionally represented in "Railway problem"

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: make railway basis right-handed

Previously the y axis was reversed!

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: move so we put the basis stuff for debugging first

No functional change.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

Revert "helixish: railway_inplane_basis should not include a scale factor"

This reverts commit 5aba1dbd37afe0078ba09dbf6cf142b559d82d2b.

OK, so we'll make the railway basis have a scale in the (x,y) plane
so that Q is at [1,0,0].  This is used for debugging output, only.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: railway_inplane_basis should not include a scale factor

Err, maybe it should ?

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output, some bodges re color handling weirdness

It seems that ring() ignores the color if it is in kw, but not
if it is a separate parameter.  WTF.

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visual: debug arrows better

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: bugfix

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging output

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

visdebug: support caller-specified colour tuples

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

helixish: debugging

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

more debugging

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>

more debugging

Signed-off-by: Ian Jackson <ijackson@chiark.greenend.org.uk>