Because we declare the functions in runtime.js we need the correct
global object to be setup otherwise they cannot be accessed when
switching to the SheetGlobalObject.
Before this commit it only allocated the global object so when it wanted
to lookup 'thisSheet' it could not find it in the global environment.
We now hotswap the global object everytime a cell evaluated.
This also fixes that SheetGlobalObject did not have an
internal_has_property meaning 'A0' could not be referenced unless it was
via a member lookup (this.A0). This was already broken before the
bindings refactoring.
The correct behavior of realms in spreadsheet is not completely clear
since what is shared between sheets is not very well defined.
The reason that just setting the SheetGlobalObject as the
global_this_value is not enough is because ECMAScript does not check the
global_this_value for members when resolving a reference in the global
environment.
This small patch allows SystemMonitor's Stack tab to show the name of
the ELF object to which the displayed address refers to. This gives a
bit more of contextual information to the viewer.
A better to show this is probably a table, but I'm not that familiar yet
with the GUI framework in general, so I'm keeping things simple.
SonarCloud flagged the read of the matches array as a potential garbage
read. I don't believe the case it flagged was possible to reach due to
how the code is structured, however we should really just be zero
initializing these stack arrays.
Some coredumps take a long time to symbolicate, so let's show a simple
window with a progress bar while they are loading.
I'm not super happy with the factoring of this feature, but it's an
absolutely kickass feature that makes crashing feel 100% more responsive
than before, since you now get GUI feedback almost immediately after a
crash occurs. :^)
Before this patch, this is what would happen after something crashed:
1. CrashDaemon finds a new coredump in /tmp
2. CrashDaemon compresses the new coredump (gzip)
3. CrashDaemon parses the uncompressed coredump and prints a backtrace
4. CrashDaemon launches CrashReporter
5. CrashReporter parses the uncompressed coredump (again)
6. CrashReporter unlinks the uncompressed coredump
7. CrashReporter displays a GUI
This was taking quite a long time when dealing with large programs
crashing (like Browser's WebContent processes.)
The new flow:
1. CrashDaemon finds a new coredump in /tmp
2. CrashDaemon mmap()'s the (uncompressed) coredump
3. CrashDaemon launches CrashReporter
4. CrashDaemon goes to sleep for 3 seconds (hack alert!)
5. CrashReporter parses the (uncompressed) coredump
6. CrashReporter unlinks the (uncompressed) coredump
7. CrashReporter displays a GUI
8. CrashDaemon wakes up (after step 4)
9. CrashDaemon compresses the coredump (gzip)
TL;DR: we no longer parse the coredumps twice, and we also prioritize
launching the CrashReporter GUI immediately when a new coredump shows
up, instead of compressing and parsing it in CrashDaemon first.
The net effect of this is that you get a backtrace on screen much
sooner. That's pretty nice. :^)
The editor now unmasks fonts on load, mapping their glyphs to the
complete unicode character set, and masks them upon saving to
reduce disk space. This is a naive approach in terms of memory
usage and can be improved but whose immediate goal is to allow
editing any glyph without concern for range allocation.
This adds an option for even more magnification, when you really need to
count pixels, as well as pausing the capture by pressing Space and
switching between magnification levels with keys 2, 4 & 8.
Previously we only had `Point::end_point_for_square_aspect_ratio`,
which was convenient for PixelPaint but assumed the aspect ratio
was always fixed at 1. This patch replaces it with a new mthod that
takes in an arbitrary aspect ratio and computes the end point based
off that.
There's some explicit casting going on in `Point.cpp` to ensure that
the types line up, since we're templating Point based on `T`.`
The previous implementation of automatic scrolling in LayerListWidget
relied on mousemove events to perform the scrolling, which didn't
produce the expected behavior. Instead use the new auto scroll timer.
Without this check we would do an unnecessary partial second round
trip because of the call chain:
LayerListWidget::set_selected_layer() ->
LayerListWidget::on_layer_select() ->
ImageEditor::set_active_layer() ->
ImageEditor::on_active_layer_change() ->
LayerListWidget::set_selected_layer()
Prior this change, custom title metrics seen in the Basalt theme
and custom icon paths in Redmond themes were dropped when saving a file.
Now any entry, even empty, will be saved. This may end up with slightly
larger files, but on other hand, users will be able to see every option
in a text editor, without a need to look at the code/docs.
This just provides some nice visual feedback when you are in the
middle of selecting a color. An eyedropper might not be the ideal
choice here, but among the cursors we already have it is the
best one.
This just allows us to ask the image editor to update the tool's
current cursor. This is useful in cases where a tool may want to
change it's cursor for some operation, and wants to invalidate the
active cursor cached in the ImageEditor.
Now we add a little marker to show the current sample position of
the CloneTool. The current implementation for updating the cursor
location is really dumb since it just updates the whole editor,
but that's a yak on the stack to fix.
There's still a lot to be desired in terms of functionality and
usability, but this is a start. When using the clone tool, you
can press Alt to sample a location, and then the brush will clone
the color from there.
Before there was dialogs to indicate if
current file is last or first.
So in this commit, I added functionality to enable
and disable forward and backward navigation based
on if the next file is avialable or not and if the
pervious file is avialable or not, respectively.
ImageViewer: Refactor code and init data
Data intialization was not happening properly, now I did
Initializations properly so that navigation can work from
the very first image.
When there is a file with the same name in the destination directory,
FileManager overwrites that file without any warning. With this change,
such a file will be automatically renamed to "emoji-2.txt", for example.
Also, currently there is a check in FileManager that makes copy and
paste of a file in the same directory no-op. This change removes that
check, because it is no longer a problem.
The problem was a bit more complex than originally anticipated,
and the root of the problem is that the "coordinates" of a pixel
are actually the top left of the pixel, and when we're really
zoomed in, the difference in editor coordinates of the top-left
and the center of the pixel is significant.
So, we need to offset the "start" point when we are painting on
the editor to account for this, based on the current scale. This
patch adds a `editor_stroke_position` in `Tool` which can be used
to compute what point (in editor coords) we should use for a given
pixel and it's corresponding stroke thickness.
Note that this doesn't really work well with the ellipse, since that
is drawn with a different mechanism. Using this new method with the
`EllipseTool` seems to give the same (or slightly worse) results, so
I have not changed anything there for now.
Previously, Any potential ImageClients would not have received an
update about the layer bitmap having been modified. This is similar
to what the other shape tools do upon completion.
Previously, we didn't ask the editor to update after drawing a
Rectangle/Line. This meant that if any part of your shape went
outside the bounds of the image, that part would not be cleared out
until the next update of the editor.
When drawing a line/rectangle/ellipse in `Tool::on_second_paint()`,
if `m_thickness * m_editor->scale()` was less than one, it would
get converted to 0 because of truncation. This was causing infinite
loops somewhere in the painter code and causing the application to
freeze.
Fixes#9986.
You could draw a Rectangle/Ellipse from the center by pressing down
the Alt key, but this was missing for lines. This commit adds it in
to keep consistency among the different shapes.
A few of the menu items were missing shortcut characters completely,
so in the interest of keyboard navigation some have been added (even
if they are not ideal).
In a few menus, severals actions had the same menu shortcut, so the
later ones were not accessible through it. These have also been
differentiated.
Previously the code assumed that the active tool had a reference to
the old editor, which is the only way we had to check if it is
active without having a reference to the editor. However, when a
tab in PixelPaint is closed, the editor is destroyed, so the `WeakPtr`
in a tool referencing the old editor is no longer valid.
This made it so that if you closed a tab, the tool would appear to be
selected in the ToolBox, but the editor would not know about it at all.
Since there's only one global toolbox, it makes sense to store the
active tool in here, since we don't really have control over the
deletion of an editor.
Previously applying filters was not calling this method, which was
not correctly triggering the `image_did_modify_bitmap` call for
the `ImageClient`s. This patch makes the filter actions call this
method.
It seems that just opening one of the Window menus triggers a repaint
of the entire editor, which is what was causing filters to update
earlier, since we were only accessing them from the menu. Using
the keyboard shortcut added in a previous commit highlighted this issue.
As pointed out by Andreas, drawing them from the bottom feels odd
since no other list in the UI is bottom-justified. The commit draws
the layers from the top of the list again.
Across the entire audio system, audio now works in 0-1 terms instead of
0-100 as before. Therefore, volume is now a double instead of an int.
The master volume of the AudioServer changes smoothly through a
FadingProperty, preventing clicks. Finally, volume computations are done
with logarithmic scaling, which is more natural for the human ear.
Note that this could be 4-5 different commits, but as they change each
other's code all the time, it makes no sense to split them up.
In a previous commit we read default values from a commit file, this
commit now also writes back any changes to those settings made by
the user. Persistent settings always feel good :^)
Someone may not want to have these things enabled by default on every
startup, but toggle them on manually. So instead of having hard-coded
everything to be enabled by default, we now query LibConfig to find
out what the preference is. These values can still always be changed
from the Menus / with shortcuts.
It's not really ideal querying LibConfig twice, but when initializing
the menu we may not have an active editor, so we need to get the
value for both the menu checkbox as well as the internal property.
This shouldn't be too much of a big deal since LibConfig caches the
values anyway :^)
The MoveTool now lets you pan if you're dragging with a right click.
Previously, right-clicking did not perform any actions at all, so this
isn't removing any old functionality.
Prior to this change, the 'Save' action was saving a file to the startup
path (or just showed a File Picker dialog) if a file has been opened
by the Open action or by drag-n-dropping a file to the program.
Prior this change, when you opened a file using the brand new Open
action and tried to change the Color Role or save it, then it would just
go back to the startup palette.
Because of the way rulers are implemented in the ImageEditor
currently, the `Fit Image To View` action doesn't work correctly
with them enabled. This patch makes them adjust to the effective
viewport area and account for the rulers.
This is a bit of a hack, but the correct way to deal with this would
be to put the rulers in a new widget so they don't interfere with
the actual viewport rect (which is being used all over).
After transitioning to FileSystemAccessServer, some of the methods
in `MainWidget` and `ProjectLoader` for opening files directly with
a filename as opposed to with a file descriptor are unused. This
commit removes them.
Use the newly added API in FileSystemAccessServer to get read-only
access to a file specified on the commandline. We no longer need to
unveil any image / .pp files on the filesystem :^)
Depending on the size / scaling of the UI, someone might want to
change what the threshold is to show the pixel grid. For instance
if you are working on a 50x50 image, and want to see the grid while
still fitting the whole image in the editor.
Since there's no UI for settings in PixelPaint right now, this
commit just uses LibConfig to read the following entry:
("PixelPaint", "PixelGrid", "Threshold")
which is then used when drawing the grid.
This removes the awkward String::replace API which was the only String
API which mutated the String and replaces it with a new immutable
version that returns a new String with the replacements applied. This
also fixes a couple of UAFs that were caused by the use of this API.
As an optimization an equivalent StringView::replace API was also added
to remove an unnecessary String allocations in the format of:
`String { view }.replace(...);`
In many cases we can lean on the name of a menu to reduce verbosity.
"View"->"Visualize/Show/Turn on/etc" is a bit redundant and clutters
the menu. It's a little thing, but it makes the system feel more
tightly integrated if we stick to the same word patterns across apps.
The editor now draws a grid showing the pixels if you are zoomed
in enough. Currently the threshold is a scale of 15 (so if one
pixel side on the image takes up > 15 pixels in the editor)
Previously we unveiled the file specified through the command-line
interface. Now, we just make the request through the
FileSystemAccessServer instead.
Previously we unveiled the file specified through the command-line
interface. Now, we just make the request through the
FileSystemAccessServer instead.
This check was fine earlier when we had access to the full filesystem,
but now that we are moving away from applications having unveiled
access to the filesystem, this check would just get rejected from
the kernel. This is rare / fast enough that performance should not
really matter in any case.
This is a feature I missed from Photoshop: it sets the scale and
position so that the image fits (it's longest dimension) into
the editor view. There's a 5% border left around the image to
provide context. This is just arbitrary seemed like the right
amount after some trial and error.
Previously EraseTool would only let you have hard lines, similar
to PenTool. After inheriting from BrushTool in previous commits,
making the eraser (optionally) behave like a brush is much easier.
We only need to change how the colors are handled for the hardness,
which is why the `draw_point()` call is a bit more involved. Just
blending the colors doesn't work here since we actually want to
replace the previous color, unlike in BrushTool where we are just
layering the color on top.
This removes all the code to handle events that was essentially
duplicated from BrushTool anyway :^)
I've also renamed "thickness"->"size" to have consistent
terminology.
Most of the logic implemented in PenTool was the same as BrushTool
anyway, with the only difference being how the actual lines were
drawn at the end. We now just override the `draw_line()` and
`draw_point()` methods instead.
We don't strictly need to override `draw_line()` here, but that
would just result in repeated calls to `draw_point()`, which is
wasteful.
Also renamed "thickness"->"size" to have consistent terminology.
The BrushTool is very cool, but it doesn't allow us to re-use any
of the code in other classes. Many of the other tools have duplicated
code for handling mouse events / keeping track of previous location,
etc.
This commit sets up BrushTool so that other tools can inherit from
it and override some virtual functions to allow similar behavior
without re-writing the code to keep track of mouse positions, etc.
In particular, we add public setters/getters for `size` and
`hardness` properties, and make `draw_point()` and `draw_line()`
virtual so that derived classes can override them.
Note: We still pass in `color` as a parameter for `draw_line()` and
`draw_point()` instead of using `color_for()` directly because it
doesn't really make sense to be constantly asking the ImageEditor
for the color when it's not really changing (for instance along all
the points of a line)
Sometimes you want to draw a straight line between 2 points, but
using the nice-looking brush we have instead of the hard line we
would get using the LineTool.
This patch adds the ability to click somewhere with the brush, and
then Shift+click somewhere else to draw a line between the two
points using the brush stroke. Seems like an obvious addition
considering we already have a helper function to draw lines :^)
The BFS implementation for BucketTool's flood-fill had sitations
which could result in infinite loop, causing OOM crashes due to
the queue growing unbounded. The way to fix this is to keep track
of the pixels we have already visited in the flood-fill algorithm
and ignore those if we ever encounter them again.
This also fixes the crashing issue from #9003. We still need a
better way to account for transparency, but that is beyond the scope
of this commit, and this issue still exists without any transparent
pixels.
This will restrict 3DFileViewer's access to the file system.
3DFileViewer loads a texture based on the path of the model loaded, this
will request access to the texture file before loading.
Previously, Assistant was able to re-execute itself due to a quirk in
the global cursor tracking logic, it seems.
Now it has become necessary to explicitly drop the lockfile before
activating the new Assistant instance.
Alternatively, this has always been a latent bug that only now has
become more likely.
Showing the position only with an active tool seems a bit confusing,
if you've opened up an image just to find out the coordinates of
a pixel for instance, there shouldn't be a need to have to select
a tool first.
Previously, we were ignoring the scale of the editor in the second
paint step. If you were zoomed in, the size while you were drawing
was not the same as the size of the final shape.
Previously, we were ignoring the scale of the editor in the second
paint step. If you were zoomed in, the size while you were drawing
was not the same as the size of the final shape.
Previously, we were ignoring the scale of the editor in the second
paint step. If you were zoomed in, the size while you were drawing
was not the same as the size of the final shape.
A previous commit I made broke layer dragging since the hole_index
was always being computed with respect to the top of the layer list
widget, however we were now drawing layers from the bottom. When
you didn't have enough layers to fill up the full height, dragging
them around would be weird.
This patch computes the hole index correctly using the same offset
we start drawing from, and fixes the behavior.
In addition to adding the action, this commit also makes the
`did_change_rect()` method take in an optional rect, which
represents the new rect position. By default it is the same as
`rect()`.
When we are cropping an image, we don't want to move the whole
cropped section to the top-left of the original image in the
ImageEditor widget, so this allows us to keep the cropped image's
position fixed.
With this patch, we now initialize the `WebContentConsoleClient` as soon
as the Page has loaded, instead of waiting for the Console Window to be
shown. This finally lets us see log messages that happened before the
window was opened! :^)
However, it is not perfect. Waiting until the page has loaded means we
lose any messages that happen *during* page load. Ideally we would
initialize the WCCC when the page *starts* loading instead, but it
requires that the page has a document and interpreter assigned and
accessible. As far as I can tell with my limited knowledge, this is not
the case until the page has completed loading.
The console widget now requests messages and receives them in bulk,
using the shiny new IPC calls. This lets it display console messages
that occurred before the widget was created. :^)
Previously, it would keep a pointer to the interpreter of the previous
page, which resulted in Crashy Fun Times.
This also changes the clearing behavior - instead of clearing the
console output every time the window is shown, we clear it when the page
changes. This is more useful, since before you would lose any log
messages that had happened before opening the window.
Having the BrowserWindow assigning Tab's internals felt a bit wrong.
This is also a step towards removing BrowserWindow as a friend class.
While I was at it, changed the Tab to having a pointer to the
ConsoleWidget instead of the Window, since that is usually what we want
to use, and it's awkward having to static_cast the main-widget
repeatedly.
When using a dark theme the name of the key does not show up well at
all. This is due to it being set with pallete().button_text(), which for
dark themes is a light color. As Keyboard Mapper does not follow the
system color theme, the text on the keys should not either.
Previously we would loop over all glyphs in the GlyphMap, compute
their rects, and then test to see if the mouse click position was
inside that rect. This is silly since each element in the glyph
map for a particular font is the same size, and we can just do
some coordinate manipulation to get the index directly.
We can now drag-and-drop files onto PixelPaint to be able to open
them. Each dropped file opens in a separate editor (which is the
default behavior of Photoshop).
Previously, all the UI setup was done in `main.cpp`, with a whole
bunch of lambdas,, and actions etc just being stored in the main
function. This is probably an artifact from back when it was first
created.
Most other applications now have a "MainWidget" class of some sort
which handles setting up all the UI/menubars, etc. More importantly,,
it also lets us handle application-wide events which we were
previously not able to do directly, since the main widget was just
a default GUI::Widget.
This patch moves all the core functionality of the PixelPaint
application into PixelPaint::MainWidget, which is then instantiated
by the main function. There is likely some more refactoring that
would help, but this commit is big enough as it is doing mostly
a direct port.
freq_bin was converted to double after it was calculated, so there was
a much higher probability it could be 0 instead of some comma number,
which meant that the bars always stayed on top.
The freq_bin in bins_per_group was multiplied only to be divided later,
which could even result in a crash if you set higher buffer size
(like 1000ms) in PlaybackManager, due to rounding errors I presume.
Casting u64 to float is probably not a safe thing to do. Also, keep
time deltas in u64 values as they can easily wrap between calculations.
This fixes CPU usage calculation when a process is spinning in a loop.
Prior this change, opening a playlist always spawned a new widget.
This could end up with having a few the same widgets, which you couldn't
even close (besides the last one).
The fd would get closed when the File went out of scope, so we couldn't
open any file specified by 'pp <path to file>'. We need the fd to be
alive and we solemnly swear to take good care of it and close it
ourselves later.
Previously all memory values on the performance was formatted as KiB,
but with such formatting it can be quite hard to read large numbers
(as mentioned by Andreas on todays office hours livestream :^)).
This patch makes use of the human readable formatting utilies and
displays them in an easier to read format.
Every tool that has a slider now registers the primary/secondary
sliders and now uses the same keyboard shortcuts to modify the
primary and secondary properties. `[` and `]` for the primary,
`{` and `}` for the secondary.
There are quite a few tools that might want to change certain values
based on consistent keyboard shortcuts. This commit allows tools to
hook up a "primary" and "secondary" slider with the base `Tool`
class, which can then handle updating those sliders with the common
shortcuts.
Note that any derived classes that want to override the `on_keydown`
function will manually need to call `Tool::on_keydown()`.
This new class will open and parse files (either images directly or .pp
project files) and one can get the parsed Image as well as other
information from it.
This patch removes a bunch of 'try_create_from..." methods from Image in
favor of using the ProjectLoader.
The only json_metadata that is available are Guides for now.
The ImageEditor knows more about the image than Image itself. So to save
a project with all the information known to the program about an image
it's logical that ImageEditor performs that task rather than the Image.
There isn't any additional data added yet, but now there's the
possibility to do so.
Our existing implementation did not check the element type of the other
pointer in the constructors and move assignment operators. This meant
that some operations that would require explicit casting on raw pointers
were done implicitly, such as:
- downcasting a base class to a derived class (e.g. `Kernel::Inode` =>
`Kernel::ProcFSDirectoryInode` in Kernel/ProcFS.cpp),
- casting to an unrelated type (e.g. `Promise<bool>` => `Promise<Empty>`
in LibIMAP/Client.cpp)
This, of course, allows gross violations of the type system, and makes
the need to type-check less obvious before downcasting. Luckily, while
adding the `static_ptr_cast`s, only two truly incorrect usages were
found; in the other instances, our casts just needed to be made
explicit.
And also try_create<T> => try_make_ref_counted<T>.
A global "create" was a bit much. The new name matches make<T> better,
which we've used for making single-owner objects since forever.
This also required adding a new hook to `ImageClient`, since there
wasn't a way of telling the ImageEditor that the full rect of the
image has changed (as when we rotate).
If we don't have enough layers to be able to scroll, the layers
are pushed to be at the top of the layer list. This doesn't make
much sense now that we are correctly drawing the layers in the
right order, so now we draw them justified towards the bottom.
Previously we were also clipping the bottom gadget slightly when
there were enough layers to scroll. Now, I'm adding some offset to
the total height to account for this and give equivalent spacing
from the top and bottom layers.
Previously the background layer was shown at the top, and layers
in front of it were shown below it. This was really unintuitive.
This patch fixes LayerListWidget to now properly differentiate
between the index of a gadget, and the index of a layer, since they
are essentially mirrored. I chose not to modify the order in which
layers are stored since back-to-front makes it really convenient
there.
This resets the DOM Inspector to a default state when re-opening it,
instead of it displaying the previous selection and properties, which
may be outdated. This is closer to how Chrome and Firefox behave.
Eventually, it probably makes sense to copy their behavior of selecting
the `<body>` element by default.
The direct-Document-access DOMTreeModel is no longer used, since the DOM
Inspector has to access the Document remotely over IPC. This commit
removes it, and renames DOMTreeJSONModel to take its place, since it no
longer has to differentiate itself from the non-JSON one.
In case that didn't make sense:
- Delete DOMTreeModel
- Rename DOMTreeJSONModel -> DOMTreeModel
Also simplify the logic by removing `Tab::view_dom_tree()`, and making
the Tab keep a pointer to the InspectorWidget instead of its Window,
since that's more often what we want to access.
We use TextEditor::on_modified_change() to update the modified window
flag, which it also works on file saves, so we don't have to unset
it there anymore!
It isn't really what the FIXME note asked about -- GUI::TextDocument
only sends us notifications about the changes, but overall I don't
think it's that bad, given that the whole window update logic is now
in one function. :^)
Previously it would only change the color of the ColorWidget itself,
but not make it the primary/secondary color. I think it feels nicer
this way, if I'm adding a color to the palette I likely want to use
it.
If you *really* need to only change the color of the palette, you
can just Ctrl+Middle click.
Previously, if you wanted to use a custom color, the only way to
do so was to first Ctrl+click on one of the pallette colors, which
would just change that palette item. Then, you would need to
manually click on that color.
Now, you can just click on the preview of the primary/secondary
color to open up the picker and only temporarily use the new color
without affecting the palette at all.
Only one place used this argument and it was to hold on to a strong ref
for the object. Since we already do that now, there's no need to keep
this argument around since this can be easily captured.
This commit contains no changes.
Without this, any resize of the GLContextWidget might leave the label
somewhere it shouldn't be. Toggling fullscreen is a single example of
that behavior.
The breadcrumbbar in here serves exactly the same purpose as the one in
File Manager. Given that, I believe it's worth to keep these two
visually consistent.
