* Split up the thirdparty code into the thirdparty folder (#3637)
* Create a new kimath static library containing all the math functions
This is part of cleaning the build system for #1906.
This is the first step to allowing non-segments in the line chain.
External routines cannot be allowed to change the line chain without
going through the internal routines. To accomplish this, we remove the
Vertex() and Point() access routines and only leave the const versions.
Transformations are given for both points as well as the chain itself.
This was re-introduced by 5d3e6e3d44
The crash happened b/c we have to manage list containers in each element
and minor adjustments cause the rest of the list to be lost. This
commit re-implements it using std::iterators and deque
Fixes#2623 | https://gitlab.com/kicad/code/kicad/issues/2623
This change completely removes the LIB_ALIAS design pattern an replaces
it by allowing LIB_PART objects to inherit from other LIB_PART objects.
The initial implementation only allows for single inheritance and only
supports the mandatory fields in the derived part because that is all
that the current symbol library file format will support. Once the new
file format is implemented and saving to the old file format is deprecated,
more complex inheritance will be added. The LIB_ALIAS information saved
in the document files was move into the LIB_PART object. This change
impacts virtually every part of the schematic and symbol library editor
code so this commit message is woefully incomplete.
REMOVE: Removed the symbol aliases concept from the schematic and symbol
editors and the symbol viewer.
NEW: Replace the symbol alias concept with simple inheritance that allows
a library symbol to be derived from another library symbol.
This update replaces the existing uses of unique pointer creation with
the C++14 std::make_unique call that provides proper memory release in
event of an exception.
The test_case_template.hpp header is deprecated in Boost, and the
latest Boost version is now throwing warnings during the build.
In Boost 1.59 (at least), this header is included transitively by the main
Boost test header, so this header is not needed by any supported Boost
version.
-Add a "wrap" option->Search results implemented as a nested list.
-Allow to go back with a "Find previous" button
-Remove the marker search option
-Fix a DLIST issue
-Add a result counter ("eg: hit(s): 1/3")
-Search history limited to 10
-Fix the search history order
-User can include or exclude references/values/texts from results
Fixes: lp:1845460
* https://bugs.launchpad.net/kicad/+bug/1845460
This means that utility programs no longer have to be manually
added to the COMBINED_UTILITY, they self-register their
information at static init time. This is basically the same concept
as the Boost test registration.
All utilities need to do now is register their UTILITY_PROGRAM info
struct with the UTILITY_REGISTRY::Register method. No headers required.
* Fix wildcard display in the file selector dialog (on GTK
it would show the regex to the user)
* Move the file extension comparison into a common function
Also rewrites the PCBNew Find dialog to make use of the above, including:
1) searching in user-defined footprint fields
2) searching in pcb text
3) a history list in the search popup
4) case sensitive searching
5) word sensitive searching
6) the ability to turn wildcard searching on/off
7) better placement of the result when the dialog obscures part of the
window
Fixes: lp:1838006
* https://bugs.launchpad.net/kicad/+bug/1838006
This changes make_lexer() so that it no longer generates a custom target
but instead attaches the generated files to an existing one (so the first
argument now is the name of an existing library or executable, and it needs
to come after the add_library/add_executable call).
The generated source is no longer listed in the project sources, as it is
added by the function. The files are generated in the build tree rather
than the source tree, and the directory is added to the include path for
the respective project as well as exported to projects linking against it.
Generated files in subdirectories are somewhat supported, but need to be
referenced with the same name as they were generated (i.e. including the
subdirectory name).
Fixes: lp:1831643
* https://bugs.launchpad.net/kicad/+bug/1831643
Fixes: lp:1832357
* https://bugs.launchpad.net/kicad/+bug/1832357
Fixes: lp:1833851
* https://bugs.launchpad.net/kicad/+bug/1833851
The QA objects link to the direct kiface objects, which creates a second
dependency on the generated lexer files. To ensure that the primary
apps are finished building them in a (potentially) different thread, we
set a false dependency on the final build product in the qa CMake.
Fixes: lp:1833851
* https://bugs.launchpad.net/kicad/+bug/1833851
The returns of KiCad file extensions on GTK contain basic regex matching
case-insensitive file extensions. We need to match these in the
provider and tests
It's a bit of a hack because they're statically initialized and
so we can't make use of the _() macro. We do still want it in the
code, however, because the string harvesting is based off of it.
Fixes: lp:1833000
* https://bugs.launchpad.net/kicad/+bug/1833000
SCOPED_TIMER is a QA-only simpler version of PROF_COUNTER.
Extend PROF_COUNTER to allow access to the std::chrono::duration
for more flexibility.
Wrap PROF_COUNTER in SCOPED_PROF_COUNTER for RAII duration
output (for the same effect as SCOPED_TIMER).
Return the EDA_DRAW_FRAME class back to common, whence it came (before it was
duplicated into legacy_wx and legacy_gal). Now there is only one
implementation (the GAL one), it doesn't need to be in a separate library.
This simplifies the dependencies for common lib users.
The dynamic ratsnest is shown while moving items but this needs to be
hidden after cancelling. Previously this was handled by the
SelectionClear event but we've changed this to (correctly) keep the
previous selection when canceling.
Fixes: lp:1831350
* https://bugs.launchpad.net/kicad/+bug/1831350
The InOutString function is not really useful, in most cases it can
be done more simple with string operator+. This function is causing
issues on MSVC (perhaps the template param names) anyway.
For some reason, older WXs on Linux (e.g. Ubuntu 16.04) would
segfault on loading the 4-tile test image. Change one tile to
a circle to fix this (presumably an old and fixed image library
function).
The bitmap pointer is not checked at copy construct. This is
an instant segfault if you copy an empty bitmap.
Fix the constructor and remove the expected test failure, from the
previous commit.
By making the caller use the "namespace" keyword, the formatter
is given the right indentation hints.
Also makes it clearer synatactically. One day, this will be a
namespace alias (needs GCC 7).
This adds a few tests on:
* LIB_PART
* SCH_PIN
* SCH_SHEET
* SCH_SHEET_PATH
These tests exercise some of the basic code paths in these classes
and show some of the expected behaviours.
None of these tests are particularly ground-breaking, but they
provide a starting point to build out further tests, and to ensure
the already-covered behaviour is stable.
It does expose some places where SCH_SHEET could probably use const.
This is done to allow access to the eeschema library
internals for purposes of test and script access, as the
DLL library has highly restrictive -fvisibility settings
that otherwise prevent the tests being able to access 99.9%
of the eeschema library functions (only a single function
is APIEXPORT'ed, therefore that's the only test we can do).
Using object libraries is a bit of a hack, and makes for
a slower link when done for multiple targets, but with the currently
supported CMake versions, it's about as good as we can get.
A better solution in the longer term may be to break eeschema_kiface(_objects)
into many smaller libraries, each of which has a much more defined scope,
rather than one big interlinked amorphous lump. This has the advantage that
each module is testable in isolation, and we get better organisation of
inter-dependencies in the codebase.
Then, the kiface DLL will gather these sub-libs and present what
is needed on the visible DLL API. Thus, we get both a testable
suite of library functions, and a restricted kiface DLL interface.
This removes the remaining hard-coded segments counts and replaces them
with the relative error calculation where the segments per arc is
determined by the maximum error we allow (smaller arcs = fewer segments)
Remove the axis numbering code from ARRAY_OPTIONS
and place in a new class, ARRAY_AXIS. This keeps
the logic for the array item numbering separate from
the logic for the array item geometry.
This simplifies the logic in the ARRAY_OPTIONS class, which
no longer has to deal with the numbering of each axis.
It was confusing that the primary frames registered their tools
differently than the other frames. In addition, since the other
frames also added their own tools, foo_actions::RegisterAllTools()
didn't really register all tool but rather those used by the
principal frame (PCB_EDIT_FRAME, SCH_EDIT_FRAME, etc.)
The angle needs to be added to the start angle.
Add tests to cover this:
* Check end point is correct
* Check midpoints are within tolerance
This would have prevented the bug fixed in 5813164d1,
as well as the bug fixed in this commit.
As well as failed tests, this manifests when routing across
Edge.Cuts arcs (unless they happen to start at 0 degrees
staritng angle).
Expose as expected failure to prove tests catch it.
Fix to follow in separate commit.
This allows CI tools to get machine-readable test reports.
By default, this is off - when you test normally, you probably
want the regular console spew.
Boost tests use the in-built Boost test command line options.
The Python tests take an --xml parameter and output the tests
there.
If neither OCC or OCE is available, the kicad2step_lib
target is not created, so the unit tests also cannot
be built. Detect missing libs and abort the test
target creation in that case.
Reverts (and fixes): 3a8ffd66c
Add a new CMake target, qa_all, which builds all
tests, tools and their deps.
Then, when KICAD_BUILD_QA_TESTS is set OFF, remove tests
and tools from the ALL target, so `make all` doesn't include
these builds.
This means even when you turn the KICAD_BUILD_QA_TESTS option
off, you still have the option to:
* Build individual tests: `make qa_pcbnew`
* Build all tests: `make qa_all_tests`
* Build all tools: `make qa_all_tools`
* Build all QA executables: `make qa_all`
This also will provide a place to hang extra logic for test routine
wrangling (e.g. by CI tools)
Update the "Compiling KiCad" dev docs.
Also, CMakeModules .cmake files should not be excluded from git
Prevent a divide-by-zero bug in SHAPE_ARC::ConvertToPolyline.
When the radius is zero, just use the initial angle (it makes
no different anyway, the result is the centre point, which is
the start point.
Add unit test of SHAPE_ARC::ConvertToPolyline.
This function has a bug when the arc is of zero radius. This
test shows the bug, but does not fix it yet.
This allows to feed any data directly into the sexpr parser. This can be
used for:
* Benchmarking, profiling and optimisation
* Fuzz testing
* Debugging
As an example, it demonstrates that a 15MB kicad PCB file (the CIAA
motherboard project) takes 1.5s to parse, and uses 140MB of heap
allocation within the Parse() method (using Valgrind massif).
This adds a test to the recent GetLayerName function, and moves
it to the base.h file for re-use for any code that needs a
layer name.
This test covers the case that caused lp:1824750 [1].
[1]: https://bugs.launchpad.net/kicad/+bug/1824750
The SEXPR class is a useful general-purpose S-Expression library
class and can (maybe) be used else where. It also should get
test coverage, as even if noone else uses it, it's critical for the
kicad2step exporter.
Also add some test coverage for some kicad2step routines. For now,
they're not useful outside kicad2step, but they are at least a useful
reference for S-Expression parsing.
On some platforms like MSVC, NaN prints as "-nan(ind)". This
is a bit needlessly ugly, so print "NaN" on all platforms
consistently.
This fixes a test failure on MSVC.
Also provide a utility function to get this path, and
a way to override at run time to quickly sub in alternative
data.
The test itself is still broken, as this plugin won't accept
a call to Load() without a KiWay.
Redesignate the eagle test as eeschema tests and build
more like the other unit tests.
Enable as a test in Ctest now that the test executes without
crashing.
The loading of the file with the hardocded part is still
not enabled, as this needs more infrastructure to support it.
Before Boost 1.64, there was no test logging function
for std::nullptr_t.
Add a logging struct to Boost to deal with this, and some
macros to assist in similar cases. These macros are bit
untidy-looking, but due to GCC bug #56480, we can't use
namespace aliasing to solve this. From Boost 1.64 onwards,
these namespaces are not needed at all.
Remove some code added to work around lack of nullptr printing
in the past.
These macros, and the nullptr printer, can be removed when
the Boost min version is 1.64 or greater.
Added checks to StrCmpNum() function in common/sting.cpp to prevent
iterating past the end of the sting. Also removed length of string
parameter since it did not seem to be used anywhere.
Fixes lp:1818157
https://bugs.launchpad.net/kicad/+bug/1818157
Add some tests of the filename splitting of WX_FILENAME objects.
Interestingly, the result of GetFuillPath on an empty path is
likely wrong as it will imply an absolute path from /. But this is
probably not a used code path. Document the current behaviour as passing
the test.
This unit tests covers the code reported by Coverity 183884 and 183894
(improper use of negative). The find_last_of can return "npos" (-1), but
it's stored as a size_t. This is correct according the WX docs,
and the substr() method it is passed to can handle npos too.
gal: pixel alignment for Cairo.
gal: pixel alignment for OpenGL
Cleanup of pixel alignment patches
gal/opengl: more pixel alignment patches:
- works on Retina displays now
- pixel alignment for circles
- correct 1-pixel strokes
- fixed unfinished 1-pixel line ends
GAL: Restore antialiasing options in Cairo
GAL: Fix arc drawing
Removes DrawArcSegment, which drew arcs whose edge was optionally
stroked (a feature that we did not use). Fixes Cairo arc drawing issue
where arcs were not visible in pcbnew.
gal: further cleanup
gal: removed unused shader parameter
When adding colors to length tuner, we depend on the window background
color, which can be set by the user, resulting in an unreadable text.
This uses the HSL conversion to detect which shade it should use for the
coloring.
Also adds test cases for round-trip HSL and HSV conversion
Fixes: lp:1814530
* https://bugs.launchpad.net/kicad/+bug/1814530
The using the "first available" numbering option, force an
artifical linear numbering scheme starting at '1'.
Start the pad name provider at the 0'th pad index.
Also adds a few related tests and adjusts tests affected by
this change to reflect that the offset still applies.
This can be expanded in future to be more general by skipping
pad (which would allow a custom start and numbering scheme
while still avoiding duplication), but for now,
this does what the UI says and avoids string changes in 5.1rc.
Fixes: lp:1814918
* https://bugs.launchpad.net/kicad/+bug/1814918
Fixes: lp:1814917
* https://bugs.launchpad.net/kicad/+bug/1814917
* Make LIB_TABLE_BASE::GetCount() return unsigned. This is more
consistent with the behaviour of STL containers (especially the
boost::ptr_vector this is really accessing). Sadly
wxGridTableBase() forces an int, so a cast is still required
at the WX interface.
* Make LIB_TABLE_BASE::At() return a reference. First, this is more
consistent with normal STL indexing operator[]'s, and secondly, it
allows an idiomatic const index method (so you can access const
LIB_TABLE_ROWs from a const LIB_TABLE_BASE).
The motivation is to allow use of this class and LIB_TABLE_ROW
in a test program, where the LIB_TABLE_BASE is const.
Some basic tests on LIB_TABLE and LIB_TABLE_ROW that demonstrate
the behaviour of fallbacks and various access methods.
Also add a few LIB_TABLE_BASE comments and changed some NULLs to
nullptr.
Libpolygon can provide its own includes via target_include_dirs PUBLIC.
This means any linking targets do not need to specifiy them manually.
As common requires polygon, the polygon dep is also now no longer
required downstream of libcommon, as it's transisitvely implied
by libcommon's target_link_libraries.
This resolves a circular dependency previously detected and also
simplifies CMakeLists.
The bitmap definitions (BITMAP_DEF and so on) do not
have any dependencies on other libs, including WX. This
means the bitmaps library can be isolated from the other
dependencies.
Common now depends on bitmaps, and libraries that depend
on common can pick it up from the common target_link_libraries,
as it is PUBLIC. This means a lot of targets no longer
need manual bitmap linkage.
This avoids a circular dependency that was previously reported
by static analysis.
Avoiding pulling in WX and other headers into the include
tree of each bitmap .cpp is a huge speed up (around 10x) in
compilation, and the generated static library is also 10x
smaller (20MB vs 200MB)
Add common as a link library to pnsrouter,connectivity.
THese library do still use common code (including bitmaps,
via base_screen.h) and this allows them to pick up the libcommon
includes correctly.
SEG collisions and distances are a fundamental part of SHAPE collision testing
and are therefore critical for DRC correctness.
Also a few Collinear tests.
This function was incorrectly processing refdeses like "U1000".
Change the algorithm to a simpler STL-compatible one and
update the tests.
Fixes: lp:1813669
* https://bugs.launchpad.net/kicad/+bug/1813669
This breaks the following functions out to a general-purposed refdes utils
header:
* MODULE::GetReferencePrefix()
* kicad_string.h RefDesStringCompare()
This acheives:
* Slimming of MODULE interface
* Placement of refdes code in common rather than pcbnew
** Testing of this code in qa_common
* Tighter and smaller includes for code that only needed refdes functions
Note: there are failing tests commited (as expected failures). These
are the cause of lp:1813669 and will be fixed as a follow-up commit.
Ian McInerney
jean-pierre charras
Seth Hillbrand
Wayne Stambaugh
Mark Roszko
John Beard
Jeff Young
Fabien Corona
Seth Hillbrand
Simon Richter
Michael Kavanagh
Tomasz Włostowski