* Fixed#501 (more Qt ownership management) - this commit contains some more changes because I had to regenerate the Qt binding sources.
* Fixed#501 (Qt object ownership transfer) - repairs, added tests
* Updated Jenkinsfile to not publish a PR build
* Update Jenkinsfile - exclude PR's from build
The reason was a synchronization issue.
Actually "LayerPropertiesNodeRef" is not a reference, but
a mirror copy of the LayerPropertiesNode the reference points
to. Changes to the original must be synchronized into the
reference and back.
This concept is a tribute to the original implementation and
the node reference was a convenience add-on to the iterator-
based API.
Better solution in general is to replace the LayerPropertiesNodeRef
concept with a real reference.
New binding: Edge#d (distance vector), Edge#clipped and Edge#clipped_line.
"intersection_point" returns nil in case of no intersection.
Documentation error fixed (Edge#distance).
New convenience functions are provided which simplify
manipulation of key bindings and menu item visibility
configuration strings. AbstractMenu#pack_key_binding
and AbstractMenu#unpack_key_binding turn a path/key
map into a single string and back. The string format
is the same than for the key-binding configuration key.
The same is provided for the menu item visibilily
with AbstractMenu#pack_menu_item_visible and
Abstract#unpack_menu_item_visible.
1.) The ExpressionContext class is a mapping of tl::Eval
and allows providing a variable context for the LQ.
Expression class is derived from ExpressionContext now.
2.) The variable lookup has been changed so that variables
can be modified even if they come from a parent context.
3.) LayoutQuery and iterator has been given an argument to
supply the context
Because long is 32bit on Windows (like int), the
conversion from long to unsigned int was subject
to sign overflow. This was fixed by going to
unsigned int via unsigned long.
- global pins have been generated for device cells too and lead
to implicit pins which may not be desired. The original problem
was how to make abstract circuits comparable. This has to be
solved differently.
- Circuit boundaries are good for displaying the boxes for
abstract circuits
In addition, this fix includes Python-related fixes: because
of the short lifetime of Python references, the functionality
was not as expected sometimes. Keeping copies of LayerPropertiesIterators
helped. Some tweaks were required to maintain the delete() semantics.
* Spice writer can now be configure to skip the debug
comments
* < and > are allowed chars in spice names now
* global net names have second prio over labels now
Spice writer: don't prefix model name with "M"
Added "device_class_mismatch" message to netlist compare
Assertion if device classes or circuits are nil on
"same_..."
- edge pairs are normalized before turning them into polygons.
This makes flat and deep implementation more consistent.
- deep region and flat regions were not cooperating in geo
checks
- unnamed layers are not registered in make_layer - this
does not make sense and will just hold a fake ref
- tests now use GDS to represent texts after transformation
(with orientation, OASIS can't do this)
- texts are more consistently handled in the tests
- test debug output is not written in the same format
than golden data unless special normalization is
requested.
- a non-orientable polygon was converted to orientable in
a text because this can be represented in GDS consistently
- DRC testsuite uses "polygons" instead of "input" to achieve
identical behavior for deep and flat mode with respect to
texts
- dbRegionTests are updated because texts are not allowed
for non-original layers too
Reimplementing virtual functions with
"const &" arguments wasn't behaving as
expected because these arguments were
copied.
Now, "const &" for arguments (in virtual
function reimplementation) is not implemented
as a copy.
In addition, now it's possible to declare
results as references always (also if const &).
See gsiTest.cc:1078 for example:
// gsi::arg_make_reference makes the function's return value
// always being taken as a reference
gsi::method<C_P, const CopyDetector &, const CopyDetector &, gsi::arg_make_reference> ("pass_cd_cref_as_ref", &C_P::pass_cd_cref)
Reimplementing virtual functions with
"const &" arguments wasn't behaving as
expected because these arguments were
copied.
Now, "const &" for arguments (in virtual
function reimplementation) is not implemented
as a copy.
In addition, now it's possible to declare
results as references always (also if const &).
See gsiTest.cc:1078 for example:
// gsi::arg_make_reference makes the function's return value
// always being taken as a reference
gsi::method<C_P, const CopyDetector &, const CopyDetector &, gsi::arg_make_reference> ("pass_cd_cref_as_ref", &C_P::pass_cd_cref)
The cause for the problem was that the layout got updated
while iterating causing the mess within the iterator.
This solution is to lock the layout while an iterator
is present. This happens for various Cell and Shapes
iterator, so it's a major enhancement.