This commit continues the work started in commits 521473ee and
e84fd464 that parallelizes certain geometric operations. This commit
cleans up the temporary arena implementations and makes them
thread-safe.
Also, in commit 521473ee, a call to FreeAllTemporary() was added
during initialization to create the heap on Windows. This is now
not necessary as the heap is created transparently on the first call
to AllocTemporary().
Commit ea6db67 added an unusual isnan macro:
#define isnan(x) (((x) != (x)) || (x > 1e11) || (x < -1e11))
Commit 8bc322e adds a preprocessor guard that looks like it would
cause the isnan function from math.h to be preferred, but doesn't
actually do that on many platforms, e.g. glibc:
#ifndef isnan
# define isnan(x) (((x) != (x)) || (x > 1e11) || (x < -1e11))
#endif
This commit renames our isnan() to make it clear that it differs
from the standard library operation, and makes it a function.
Fixes#603.
We use std::fabs, but <math.h> doesn't provide it; it worked by accident.
Include <cmath> that provides std::fabs, and switch all other headers to
their C++ variants, too.
* Don't use a reserved identifier in include guards.
* Use fabs() from <cmath> instead of our own ffabs().
This shouldn't make any difference with modern toolchains.
* Convert a few preprocessor macros to constexprs.
After this commit, dbp() is renamed to DebugPrint() and moved to
platform.cpp, next to other similar functions. The existing short
name is provided by a preprocessor macro, similar to ssassert().
This leaves just the (rather hacky) temporary heap in util*.cpp.
This commit performs three related cleanups:
* The slvs library no longer uses explicit platform initialization
(which drags in the side effects of InitPlatform that are not
desirable in a library). Instead, it just ensures that it has
the temporary heap, which is what it was callingInitPlatform for.
* InitPlatform is simplified and moved to platform.cpp, next to
other path related functions.
* InitPlatform is renamed to InitCli and is called from InitGui
implementations. GUI toolkits sometimes have options they use
internally (that's the case for for GTK and Cocoa at least),
and we shouldn't try to parse those as a file to open.
Historically SolveSpace used its own heap on Windows since it gave
better control and debugging options, but a lot of development these
days happens on Linux, where that heap was a stub around malloc/free,
and also Windows debugging tools got a lot better.
In terms of immediate benefit, this commit fixes heap corruption
on Windows introduced in commits b4e1ce44 and 47e82798, caused
by allocating with HEAP_NO_SERIALIZE in parallel from OpenMP threads.
Without HEAP_NO_SERIALIZE there's no performance benefit to keeping
our own heap, either.
The vl() function is also removed because for development there are
better tools now, and the only place where it was permanently called
from became a no-op, since temporary heap always validates after
FreeAllTemporary() recreates it.
See:
https://github.com/solvespace/solvespace/issues/92#issuecomment-567831112
The problem was first introduced here:
dabd57847e
and later "fixed" here:
f324477dd0
by setting the stack size to /STACK:33554432
Solvespace now starts up even with /STACK:554432
According to this:
https://en.cppreference.com/w/cpp/language/value_initialization
```
"2) if T is a class type with a default constructor that is neither user-provided nor deleted (that is, it may be a class with an implicitly-defined or defaulted default constructor), the object is zero-initialized and then it is default-initialized if it has a non-trivial default constructor; "
```
So removing the `{}` should leave both the `System` and `TextWindow` class instances properly initialized.
It makes no sense to solve by substitution (therefore weakening rank
check) in SolveRank(), since that's the whole point of SolveRank().
In addition, because SolveRank() is currently always called right
after AddConstraint(), forceDofCheck would always be true anyway.
In addition, it makes no sense to have TestRankForGroup() dependent
on the result of the previous solve. (For SolveGroup(), solving by
substitution after we know that rank test succeeds makes dragging
points much faster.)
Supported metric units: km, m, cm, mm, µm, nm.
Supported USCS units: in, mil, µin.
Also, use the newly introduced unit formatting machinery in tools for
measuring perimeter, area and volume, so that e.g. volume is not
displayed in millions of cubic millimeters.
Its only use was in a context where it was completely equivalent to
MemFree, so just use that instead, and keep the temporary heap as
purely an arena allocator, that could use something like bump
pointer.
This is useful in niche cases, like making angular measurement tools.
Also, use simpler and more principled code for numeric precision
while editing constraints: don't special-case angles, but use up to
10 digits after the decimal point for everything.
We plan to use flatbuffers in the future for the next generation of
the .slvs file format, so flatbuffers are built unconditionally; and
the Q3DO exporter itself is tiny.
Modifying the original entities instead of deleting them, retains the
original associated constraints. This makes creating rounded rectangles
a lot easier.
This serves two purposes.
First, we want to (some day) convert these messages into a less
obtrustive form, something like toaster notifications, such that they
don't interrupt workflow as harshly. That would, of course, be
nonblocking.
Second, some platforms, like Emscripten, do not support nested event
loops, and it's not possible to display a modal dialog on them
synchronously.
When making this commit, I've reviewed all Error() and Message()
calls to ensure that only some of the following is true for all
of them:
* The call is followed a break or return statement that exits
an UI entry point (e.g. an MenuX function);
* The call is followed by cleanup (in fact, in this case the new
behavior is better, since even with a synchronous modal dialog
we have to be reentrant);
* The message is an informational message only and nothing
unexpected will happen if the operation proceeds in background.
In general, all Error() calls already satisfied the above conditions,
although in some cases I changed control flow aroudn them to more
clearly show that. The Message() calls that didn't satisfy these
conditions were reworked into an asynchronous form.
There are three explicit RunModal() calls left that need to be
reworked into an async form.
This commit removes Platform::Window::Redraw function, and rewrites
its uses to run on timer events. Most UI toolkits have obscure issues
with recursive event handling loops, and Emscripten is purely event-
driven and cannot handle imperative redraws at all.
As a part of this change, the Platform::Timer::WindUp function
is split into three to make the interpretation of its argument
less magical. The new functions are RunAfter (a regular timeout,
setTimeout in browser terms), RunAfterNextFrame (an animation
request, requestAnimationFrame in browser terms), and
RunAfterProcessingEvents (a request to run something after all
events for the current frame are processed, used for coalescing
expensive operations in face of input event queues).
This commit changes two uses of Redraw(): the AnimateOnto() and
ScreenStepDimGo() functions. The latter was actually broken in that
on small sketches, it would run very quickly and not animate
the dimension change at all; this has been fixed.
While we're at it, get rid of unused Platform::Window::NativePtr
function as well.
This is to address MSVC warnings.
This commit changes a few configuration fields to use double instead
of float. There doesn't seem to be any reason these use float except
for the legacy Windows code using float for saved configuration.
Changing their type to double improves consistency.
This commit merges all ad-hoc file dialog code, such as the feature
where dialogs remember last location and format, and exposes it
through a common interface.
This commit also significantly improves Gtk dialog handling code.
This commit changes the awfully specific code for dialogs with
messages duplicated three times to go through a generic interface.
It also fixes some issues with the way translated messages
were parameterized.
This commit removes the custom message dialog box used on Windows,
for several reasons. First, it was the last element not respecting
HiDPI displays. Second, other OSes do not easily provide this much
control over rendering default message boxes, and both Gnome and
macOS frown upon non-standard renderings such as those; so the custom
rendering was already not used on the other OSes.
This commit mostly just changes the settings code to be in line with
the rest of the platform abstractions, although it also fixes some
settings names to be consistent with others, and uses native bool
types where applicable.
This commit also makes settings-related operations much less
wasteful, not that it should matter.
This commit removes a large amount of code partially duplicated
between the text and the graphics windows, and opens the path to
having more than one model window on screen at any given time,
as well as simplifies platform work.
This commit also adds complete support for High-DPI device pixel
ratio. It adds support for font scale factor (a fractional factor
on top of integral device pixel ratio) on the platform side, but not
on the application side.
This commit also adds error checking to all Windows API calls
(within the abstracted code) and fixes a significant number of
misuses and non-future-proof uses of Windows API.
This commit also makes uses of Windows API idiomatic, e.g. using
the built-in vertical scroll bar, native tooltips, control
subclassing instead of hooks in the global dispatch loop, and so on.
It reinstates tooltip support and removes menu-related hacks.
This commit removes a large amount of redundant code that needed
to be kept in sync between platforms and also makes it much easier
to add new menu-related functionality since little to no platform
code needs to be altered anymore.
This commit also greatly improves code locality in context menu
handling by allowing context menu click handlers to be closures.
This commit temporarily introduces a SetMainMenu API, which is rather
hacky but only necessary until an abstraction for windows is added.
This changes the assertion failure behavior to be the same in debug
and release builds: to show the complete failure message, and
to offer to restart the application or defer to Windows Error
Reporting to generate a backtrace. Contrary to popular belief,
WER is not useless, and since SolveSpace publishes pdb files,
WER-generated reports can be symbolized.
This commit also addresses the long-standing problem where showing
a dialog on fatal error would re-enter the application code, thus
causing another error or a crash that is more fatal than the current
one.
According to the C standard all preprocessor definitions starting
with an underscore are reserved for standard and implementation use,
so don't use those. Also, sort and unique include directives.
This commit updates a *lot* of rather questionable path handling
logic to be robust. Specifically:
* All path operations go through Platform::Path.
* All ad-hoc path handling functions are removed, together with
PATH_SEP. This removes code that was in platform-independent
parts, but had platform-dependent behavior.
* Group::linkFileRel is removed; only an absolute path is stored
in Group::linkFile. However, only Group::linkFileRel is saved,
with the relative path calculated on the fly, from the filename
passed into SaveToFile. This eliminates dependence on global
state, and makes it unnecessary to have separare code paths
for saved and not yet saved files.
* In a departure from previous practice, functions with
platform-independent code but platform-dependent behavior
are all grouped under platform/. This makes it easy to grep
for functions with platform-dependent behavior.
* Similarly, new (GUI-independent) code for all platforms is added
in the same platform.cpp file, guarded with #ifs. It turns out
that implementations for different platforms had a lot of shared
code that tended to go out of sync.
The existing code is horrible and needlessly platform-dependent.
Even worse, it causes a freeze on GTK. Instead of propping that up
with a few more crutches, just fix the root cause.
This is a fairly standard CAD feature; it conveys the same
information and has the same recovery path, without erroring out,
so seems like an obvious win.
