+ Implement sRGB to Linear and Linear to sRGB conversions.
+ Fix an issue in the AntiAliasing adaptive logic.
+ Improves the ambient color use on the shader.
+ Improves post-shader calculation. Fixes the saturated dark regions.
+ Improves the transparency of soldermak material.
+re-implement anti-aliasing, making it more acurate, not blured, better
and much faster (unsing much less lines of code)
+rewrite some function on camera and raypacket.
+small improvements on raytracing realtime, remove useless ifs
comparisons.
Use the correct math to unproject the mouse movement from screen space
to the Z=0 plane on the camera space. This handles both perspective
and orthographic projection at any window size or zoom level.
OpenMP does not yet permit unsigned loop counters. Most compilers seem
to allow them anyway, but a few have been seen to complain.
Mistake introduced in:
commit 82ed0fde32
Author: Chris Pavlina <pavlina.chris@gmail.com>
Date: Sun Aug 28 02:05:49 2016 -0400
Fix shared data access in raytracer, tidy up render loop
Render loop lost track of which blocks had been rendered due to a
synchronization issue. Specifically, std::vector<bool> is special-cased
as a bitfield and so is pretty much guaranteed not to be atomic unless
synchronized externally.
Also:
- Clean up types of a few variables
- Clean up openmp sharing type of variables (may result in better
optimization)
- Replace shared rendered block count with an openmp reduction (results
in fewer locks)
Fixes: lp:1608289
* https://bugs.launchpad.net/kicad/+bug/1608289
* Split and rewrite the preview window and canvas.
* Create a new class for handling the board information.
* Adds new render targets: openGL, legacy, and ray tracing.
* Render targets take full advantage of the new 3D plugins system and 3D cache
for a fast 3D model loading.
* Faster board loading.
* New OpenGL render is faster than the old one.
* New ray tracing render target with a post processing shader.
* Use of new 3D plugins (WRL, X3D, STEP and IGES) and 3D model caching.
* Preview of 3D model while browsing the file name.
* 3D preview of the footprint while adding / align 3D shapes.
* Render of 3D models according to attributes: Normal, Normal+Insert, Virtual.
* Pivot rotation centered in one point over the PCB board.
* Shortcuts keys improved for XYZ orientation..
* Animated camera.
* Use platform cache path insted of the KiCad configuration path to write the
3D model cache.
* OSX: ${HOME}/Library/Caches/kicad/3d
* Linux: ${XDG_CACHE_HOME}/kicad/3d or ${HOME}/.cache/kicad/3d
* Windows: AppData\Local\kicad\3d
* Add a Help button and text to the 3D alias configuration dialog
* Display KISYS3DMOD and internally defined KiCad path env vars
* Add a button to bring up the "Configure Paths" (env vars) dialog
* Remove the file name mapping feature from the resolver; this results in
the resolver always using the current definition rather than a possibly
outdated cache value.
Array is in pairs but has a single sentinel, trips up Coverity (CID 147346) and
does allow an out-of-bounds write if WX_GL_SAMPLES were ever removed from the
array.
* Paths defined via pcbnew->Preferences->Configure Paths are now included in
the 3D file browser's drop-down list except for KICAD_PTEMPLATES, KIGITHUB,
KISYSMOD, any paths resembling a URL, and any non-existent paths. When an
absolute path is shortened using one of these path aliases, let's say
'KISYS3DMOD' then the name appears in the pcb file as "${KISYS3DMOD}/blah.wrl".
If a user defines and actual environment variable with the same name, then
that user defined variable will have precedence over the internally defined
variable.
* Paths relative to the current project directory are no longer represented
as "some/path/to/file.wrl". For compatibility with previous KiCad versions
it shall always be stored as ${KIPRJMOD}/some/path/to/file.wrl.
* Various restrictions on alias path values have been removed since the
imposed restrictions could result in corrupted user data as aliases are
silently dropped.
* Overall the patch should improve back-compatibility by providing a ${ENV_VAR}
mechanism for supporting different model root directories and ensuring that
paths relative to KIPRJMOD are expressed in a way that is compatible with
earlier KiCad versions. This allows users more flexibility and the ability
to work in a way that ensures 3D model files can be correctly resolved in
earlier versions of KiCad. Users who do not care about back-compatibility
may prefer to use the newer alias system.
The attached patch fixes a segfault due to a race condition.
If a user starts eeschema with an empty sheet and clicks
the 'cvpcb' button, memory is corrupted and the program
segfaults. The issue appears to arise from multiple threads
accessing the 3D cache and resolver. This patch makes
relevant code thread-safe by using a wxCriticalSection.
(In this case the schematic frame exists, but some members are not initialized as expected, and a crash happens when calling the schematic editor dtor)
Minor other fix: avoid multiple very similar strings to translate in 3d_filename_resolver.cpp
The attached patch ensures the S3D_MASTER class does not cause
a segfault in pcbnew when used from Python scripting.
The issue is due to an "extern KIWAY* TheKiway" which was exposed
to allow S3D_MASTER to use the new resolver to handle 3D filenames.
Unfortunately TheKiway cannot be assigned a value until a window is
created which implements Kiway. The change introduced by this
patch should not alter the behavior of S3D_MASTER compared to
the behavior before the 3D merge.
The extern variable is only there to support S3D_MASTER which in
turn is there to support the current 3DViewer. In the planned software
structure the rather complex S3D_MASTER class is replaced with
a simple struct which only holds the model name and position/
orientation/scale data as stored in the PCB file; since the replacement
does not perform filename resolution there is no need to expose the
Kiway. For example in Mario's branch we removed the extern and
S3D_MASTER class many months ago.
* Dialog copper zones: Always enable thermal shape settings, because even with no thermal, some pads can use thermal option in local pad settings.
* GAL mode: shows page limits in gray color, like in legacy mode, and according to comments in sources.
Including the inline of the Output directory and its text field/browse button.
Moving the plot format to the left of the plot options as the plot options depend on the plot format.
non-transformed S3DMODEL for rendering.
+ S3D_CACHE now caches S3DMODEL and returns a pointer to a single
instance of the S3DMODEL for repeated calls to Prepare()
+ Beginning the VRML Plugin
In polygon calculations (combining polygons, fracture) the mode of calculation (fast or strictly simple polygon option) as no more a default value, because choosing the best mode is better to optimize the calculation time.
'GetVersion'.
+ Introduced normal calculations. This is buggy due to multiple equal
face normals distorting the vertex normal; this will be fixed by
removing the redundant face normals from the calculations.
* Rename Eda_FileSelector to EDA_FILE_SELECTOR.
* Add optional pointer to wxString to save the most recently used path to EDA_FILE_SELECTOR.
* Rename Eda_DirectorySelector to EDA_PATH_SELECTOR.
* Replace wxGetCwd() with wxStandardPaths::GetDocumentsDir(). This fixes the windows issue
where wxGetCwd() returns the path where the executable files are located.
* Add code to handle most recently used path to EDA_BASE_FRAME.
* Fix string formatting error in kicad/files-io.cpp.
* Remove setting and restoring current working directory when launching GerbView. Setting the CWD
has no effect on the launched executable which starts with it's own CWD.
* Allow project path to be passed to GerbView when launched from KiCad.
* Note: this is a work in progress. Do not expect every path and/or file selection dialog to properly
update the most recently used path. The correct solution to this problem requires a much more
well though out solution which will not happen until after the next stable release.
PAD_SHAPE_T and PAD_ATTR_T still have a double definition (new names and old names) to be sure python scripts are nor broken by the change.
PAD_DRILL_SHAPE_T does not have a double definition, because it is unlikely oblong holes are used in python scripts.
Double definitions will be removed in the (next) future.
In 3D viewer, this option can take a *very long* calculation time (minutes instead of seconds, or hangs) and they are now using aFastMode = true to do polygon calculations, especially for zones inside zones belong to the same net but having different priority (which creates weak polygons).
aFastMode = false is the default for other calculations (zone filling) as before.
* Set initial path to the current project path.
* Remember user selected path and file name between saves.
* Prompt user when over writing existing files.
Fix some issues in filling zones normals.
Fix an issue with some models that have materials but didn't defined the diffuse color.
Workaround for Bug #1443431.
Implement some missing "code logic" for pervertexperface normals.
Remove some not used functions.
Calculate normals using double type.
Fix minor 3D viewer issues (includling change number of segments and its optimization to draw pad holes.)
Fix Bug #1439132 (track and via list on toolbar not always refresh after loading a board).
Remove useless tool in Modedit.
* Make OK button the default action in the 3D viewer option dialog per UI
policy.
* Fix standard button sizer padding in 3D viewer options dialog.
* Let sizers do their job by using default size for 3D viewer options
dialog.
* Set 3D viewer hot key to Alt+3 in all parent windows where 3D viewer is
accessible.
* Minor code cleaning and coding policy fixes.
Very minor fixes: remove duplicate #define in vrml_v2_modelparser.cpp, and "undo" a very minor change I made in class_board.cpp when I tried to fix a Pcbnew crash.
Ensure some values initialized by the schematic editor and used by libedit are also initialized in libedit even if the schematic editor was not started.
Remove useless commands (save/load project config) in libedit, because now they have no sense.
wxUpdateUIEvent events can be sent to parent frames, when opening a menu in a child frame, if parent and child frame share same ID fro menuitems (or tools)
The wrong menuitem can be used in some cases ( because there are more than one menuitem with the same identifier), by a wxUpdateUIEvent event function run in a parent frame.
3dviewer: back to double (from float) in class S3D_MASTER for 3 members (m_MatScale, m_MatRotation, m_MatPosition) which are used in dialogs and r/w file functions, which expect double.
Using float create minor but unwanted issues in r/w file functions.
S3D_MATERIAL: enable all color options: if a 3d shape has bad color parameters, the shape must be modified, not the 3d rendering.
The rendering now matches what we see in other vrml viewer like FreeCAD.
Some minor coding style fixes.
2) Change from legacy Cu stack to counting down from top=(F_Cu or 0).
The old Cu stack required knowing the count of Cu layers to make
sense of the layer number when converting to many exported file types.
The new Cu stack is more commonly used, although ours still gives
B_Cu a fixed number.
3) Introduce class LSET and enum LAYER_ID.
4) Change *.kicad_pcb file format version to 4 from 3.
5) Change fixed names Inner1_Cu-Inner14_Cu to In1_Cu-In30_Cu and their
meanings are typically flipped.
6) Moved the #define LAYER_N_* stuff into legacy_plugin.cpp where they
can die a quiet death, and switch to enum LAYER_ID symbols throughout.
7) Removed the LEGACY_PLUGIN::Save() and FootprintSave() functions.
You will need to convert to the format immediately, *.kicad_pcb and
*.kicad_mod (=pretty) since legacy format was never going to know
about 32 Cu layers and additional technical layers and the reversed Cu
stack.
This workaround was already used (bzr 4301, in file plot_board_layers.cpp) also to avoid similar crashes (uses 18 segments instead of 16 to approximate a circle by segments).
Minor code cleanup in dialog_plot.
- Converted the Next/Prev C casts to static casts and removed the type
unsafe ones
- Splitted as virtual the VIA::Flip member instead of using RTTI
- Heavily refactored the 'unconnected track' cleanup routine
- Misc constification
- SEGVIA becomes VIA
- Drill size moved from TRACK to VIA
- Removed shape from TRACK, becomes ViaType in VIA
- GetTrace becomes GetTrack, for uniformity
- Some minor constification and typo fixes
Eeschema, Pcbnew, and Cvpcb all run under the same process now,
FOR THE VERY FIRST TIME!
*) Added KIWAY::PlayerCreate(), PlayerClose(), and PlayersClose().
*) Factored FRAME_T into <frame_type.h> from ID_DRAWFRAME_TYPE.
*) Found that the following command line is helpful for collecting all the *.kiface
files into the <build>/kicad/ directory so that kicad can find them.
$ cp `find . -name '*.kiface'` kicad/
Maybe somebody will want to rework how the CMake files are organized so all
the binaries can go into the same place. See python-a-mingw-us.
*) This might fix the problem on the Mac where child process windows were not
coming to the front. See ->Raise() in kicad/mainframe.cpp.
*) You can set USE_KIFACE to 0 in kicad/mainframe.cpp to chain load child exes
instead of using the KIFACE modules directly, i.e. revert.
! The initial testing of this commit should be done using a Debug build so that
all the wxASSERT()s are enabled. Also, be sure and keep enabled the
USE_KIWAY_DLLs option. The tree won't likely build without it. Turning it
off is senseless anyways. If you want stable code, go back to a prior version,
the one tagged with "stable".
* Relocate all functionality out of the wxApp derivative into more finely
targeted purposes:
a) DLL/DSO specific
b) PROJECT specific
c) EXE or process specific
d) configuration file specific data
e) configuration file manipulations functions.
All of this functionality was blended into an extremely large wxApp derivative
and that was incompatible with the desire to support multiple concurrently
loaded DLL/DSO's ("KIFACE")s and multiple concurrently open projects.
An amazing amount of organization come from simply sorting each bit of
functionality into the proper box.
* Switch to wxConfigBase from wxConfig everywhere except instantiation.
* Add classes KIWAY, KIFACE, KIFACE_I, SEARCH_STACK, PGM_BASE, PGM_KICAD,
PGM_SINGLE_TOP,
* Remove "Return" prefix on many function names.
* Remove obvious comments from CMakeLists.txt files, and from else() and endif()s.
* Fix building boost for use in a DSO on linux.
* Remove some of the assumptions in the CMakeLists.txt files that windows had
to be the host platform when building windows binaries.
* Reduce the number of wxStrings being constructed at program load time via
static construction.
* Pass wxConfigBase* to all SaveSettings() and LoadSettings() functions so that
these functions are useful even when the wxConfigBase comes from another
source, as is the case in the KICAD_MANAGER_FRAME.
* Move the setting of the KIPRJMOD environment variable into class PROJECT,
so that it can be moved into a project variable soon, and out of FP_LIB_TABLE.
* Add the KIWAY_PLAYER which is associated with a particular PROJECT, and all
its child wxFrames and wxDialogs now have a Kiway() member function which
returns a KIWAY& that that window tree branch is in support of. This is like
wxWindows DNA in that child windows get this member with proper value at time
of construction.
* Anticipate some of the needs for milestones B) and C) and make code
adjustments now in an effort to reduce work in those milestones.
* No testing has been done for python scripting, since milestone C) has that
being largely reworked and re-thought-out.
Fix also other very minor issues.
Realistic mode shows or not the copper thickness (depending on selected option in preferences) to speed up the screen redraw.
Add missing calls to Show( false ) to some main frames (Kicad, eeschema, gerbview), to force config saving when closing these windows.
Code cleanup and minor coding style fixes in footprint_wizard_frame
Now footprints, after loaded by reading a netlist are grouped by sheets by the footprints spread function, and the grouping is better.
Rename 2 files. Fix minor issues. Clean code
inside the main board outlines.
specctra_export.cpp: always approximate arcs in outlines by 36 segm for 360 degrees,
even for arc angle which differs from 90 degrees.
90 degrees is the default when creating an arc, but the arc value can be edited and have any value between 0 and 360 degrees.
Previously, 9 segments were used regardless the arc angle.
File "rules" has instructional text as comments near top.
2) Convert all text files in repo to LF line ending form.
Any checkout done with "rules" in play will convert the working
tree to native line ending, while keeping repo as LF line ending.
The plan goes like this:
- eeschema still uses int in decidegrees
- all the other things internally use double in decidegrees (or radians
in temporaries)
- in pcbnew UI the unit is *still* int in decidegrees
The idea is to have better precision everywhere while keeping the user with int i
angles. Hopefully, if a fractional angle doesn't come in from the outside, everything
should *look* like an integer angle (unless I forgot something and it broke)
When the time comes, simply updating the UI for allowing doubles from the user should
be enough to get arbitrary angles in pcbnew.
* All items shapes are converted to polygons.
* Polygons are merged layer by layer (for calculation time reasons,zones are not merged)
* for copper layers, vias and pads holes are substracted from polygons (but, for calculation time reasons, not inside zones areas).
* the look is better, mainly when displaying the copper thickness
* solder and paste layers are now shown in 3D viewer.
* the code was seriously cleaned (but still needs to be enhanced).
* Note this is a work in progress which needs refinements.
- Removed spurious int casts (these are truncated anyway and will break
doubles)
- Applied the Distance, GetLineLength, EuclideanNorm, DEG2RAD, RAD2DEG
ArcTangente and NORMALIZE* functions where possible
- ArcTangente now returns double and handles the 0,0 case like atan2, so
it's no longer necessary to check for it before calling
- Small functions in trigo moved as inline
* Create separate NETLIST object to hold contents of netlist files.
* Read entire netlist and footprint link files before making applying
changes to board.
* Add BOARD::ReplaceNetlist() function to eliminate the calls between the
NETLIST_READER, PCB_EDIT_FRAME, and BOARD objects.
* Change placement of new components below the center of the current board
or in the center of the page if the BOARD is empty.
* Add dry run option to netlist dialog to print changes to message control
without making changes.
* Add button to netlist dialog to allow saving contents of message control
to a file.
* Eliminate the need to compile netlist_reader_*.cpp in both CvPcb and Pcbnew.
* Add netlist_reader_*.cpp to the pcbcommon library.
* Remove redundant load component link file code from CvPcb.
* Modify CvPcb new to work with the new NETLIST_READER object.
* Add compare() function and < and == operators to FPID object.
* Add REPORTER class to hide an underlying string writing implementation for
use in low level objects. Thank you Dick for the idea.
* Lots of minor coding policy, Doxygen comment, and missing license fixes.
Pcbnew, layer manager: add option (popup menu) to always keep copper layers not visible but the active layer, even when the active layer is changed.
Usefull for multilayer ( more than 4 layers) PCBs.
In particular the new mechanism for handling extended color palettes is in place,
included renaming the ini keys and saving the color name instead of its index; this means better forward compatibility with palette changes.
Since ini keys are changed, colors will be reset
* Complete encapsulation of the ZONE_CONTAINER class.
* Fixed a bug in the EDA_TEXT::Format() default effects behavior due to
improper conversion to internal units and missing test for default
thickness.
* Improved grammar of some Pcbnew context menu strings.
* Moved some more cleverly hidden BOARD object methods into class_board.cpp.
* Lots of coding policy and documentation fixes.
* Complete encapsulation of the MODULE class.
* Complete encapsulation of the EDA_TEXT class.
* Encapsulate most of the ZONE_CONTAINER class.
* Add pcbcommon library as a dependency for reSWIGging the scripting
support. This should cover most dependency cases.
* Encapsulate most of the MODULE class.
* Start encapsulating the DIMENSION class.
* Lay some groundwork for EDA_TEXT encapsulation.
* Move cleverly hidden MODULE functions into class_module.cpp.
* Use std::swap to exchange TEXTE_PCB values for undo/redo.
* Remove unused members from MODULE class.
* The usual coding policy and documentation fixes.
and therefore tracks are now dragged when a end point is inside a pad, not necessary on the pad position.
However, drag functions still need more cleanup.
* serious code cleanup (remove duplicate code)
* shows (option in 3D preference menu) the copper items (tracks, zones...) in 3D, using 35 microns copper thickness.
However, because there are a lot more3D data to show (roughly 4 times more), this is slower.
All: press ctrl+shift key when moving the mouse allows the graphic cursor to be moved outside the grid.
(useful to place graphic objects, texts in any position, regardless the current grid)
* Add s-expression file parser object and keyword files.
* Fix minor issues with s-expression file formatting.
* Fix a minor bug the zone container fill state parsing in the legacy plugin.
* Move EDA_TEXT visibility definition to eda_text.h.
* Add minor BOARD_ITEM object improvements to support s-expression file
parser.
// This provides better project control over rounding to int from double
// than wxRound() did. This scheme provides better logging in Debug builds
// and it provides for compile time calculation of constants.
#include <stdio.h>
#include <assert.h>
#include <limits.h>
//-----<KiROUND KIT>------------------------------------------------------------
/**
* KiROUND
* rounds a floating point number to an int using
* "round halfway cases away from zero".
* In Debug build an assert fires if will not fit into an int.
*/
#if defined( DEBUG )
// DEBUG: a macro to capture line and file, then calls this inline
static inline int KiRound( double v, int line, const char* filename )
{
v = v < 0 ? v - 0.5 : v + 0.5;
if( v > INT_MAX + 0.5 )
{
printf( "%s: in file %s on line %d, val: %.16g too ' > 0 ' for int\n", __FUNCTION__, filename, line, v );
}
else if( v < INT_MIN - 0.5 )
{
printf( "%s: in file %s on line %d, val: %.16g too ' < 0 ' for int\n", __FUNCTION__, filename, line, v );
}
return int( v );
}
#define KiROUND( v ) KiRound( v, __LINE__, __FILE__ )
#else
// RELEASE: a macro so compile can pre-compute constants.
#define KiROUND( v ) int( (v) < 0 ? (v) - 0.5 : (v) + 0.5 )
#endif
//-----</KiROUND KIT>-----------------------------------------------------------
// Only a macro is compile time calculated, an inline function causes a static constructor
// in a situation like this.
// Therefore the Release build is best done with a MACRO not an inline function.
int Computed = KiROUND( 14.3 * 8 );
int main( int argc, char** argv )
{
for( double d = double(INT_MAX)-1; d < double(INT_MAX)+8; d += 2.0 )
{
int i = KiROUND( d );
printf( "t: %d %.16g\n", i, d );
}
return 0;
}
* BLOCK_SELECTOR class is not longer derived from EDA_ITEM.
* Encapsulate BLOCK_SELECTOR class member variables and add access methods.
* Move HandleBlockBegin() function from block_commande.cpp to drawframe.cpp.
* Remove virtual methods from top level derived objects per future
coding policy change.
* Remove Doxygen copydoc statement from objects derived from EDA_ITEM
since the comments are automatically copied to the derived object.
* Removed copy and pasted Doxygen comments from objects derived from
EDA_ITEM.
jean-pierre charras
Chris Pavlina
Mario Luzeiro
Simon Richter
Cirilo Bernardo
José Ignacio Romero
Wayne Stambaugh
Jon Neal
Maciej Suminski
Wayne Stambaugh
Matthew Petroff
Cirilo Bernardo
Bernhard Stegmaier
Cirilo Bernardo
unknown
unknown
Kristian Nielsen
Mario Luzeiro
Tomasz Wlostowski
unknown
unknown
Garth Corral
Brian Sidebotham
Mark Roszko
Sven D
Camille
Henri Valta
Andrew Zonenberg
unknown
Dick Hollenbeck
Lorenzo Marcantonio
unknown
Heikki Pulkkinen
marco.
Marco Mattila