| Commit message (Collapse) | Author | Age |
|---|
| |
|
|
|
| |
This includes a fix to prevent the cursor from being marked as inactive
while at least one mouse button is still being held.
|
| |
|
|
|
|
|
| |
The metadata specification has changed to use '/' as the separator
between the program name and program version in the name string, and
each author line is now no longer followed by the year in which the
author last contributed to the program.
|
| |
|
|
|
|
|
|
|
|
| |
When opening the previously-visited directory in the file device, the
directory data is read from a cache, preventing the need to have it
regenerated from scratch. The directory data includes the selected
child, which meant that instead of child 0 being selected as per the
specification, the previously-selected child was selected instead.
To fix this, the child is deselected as the directory data is cached.
|
| |
|
|
|
| |
This was erroneously set to 0 from before the memory device was
implemented.
|
| |
|
|
|
|
|
|
| |
In programs where the same sprite is drawn many times repeatedly, a lot
of time is saved by caching the transformed sprite data instead of
having to recalculate it for every draw operation. No testing has been
done on the efficiency improvements this offers, but it doesn't seem
like it could have any downsides.
|
| |
|
|
|
| |
Writing a value to the pointer and length ports wasn't working, because
the cached write was never getting committed.
|
| |
|
|
|
|
|
|
|
|
|
| |
An overflow error was causing the line drawing method to loop forever
any time a line of length 0x4000 (16384) or longer was drawn. The issue
was occurring because both e1 and dx (or dy) would have a value of at
least 0x4000, and so on lines 291/292 the sum would exceed 0x8000, the
maximum value of an i16. The value would then wrap and break the
assumptions of the line drawing algorithm, causing it to loop forever.
This was fixed by increasing the size of the affected types.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
When determining whether or not to draw each pixel of a textured line
or rectangle, the modulo-by-eight operation was being performed on a
signed value, which was returning a negative value when the pixel being
drawn was off-screen. When the negative value was then converted to an
unsigned value, the result was close to usize::MAX, and was causing an
out-of-bounds array access.
To fix this, the value is converted to an unsigned value before taking
the modulo.
|
| |
|
|
|
|
|
|
| |
File and stream data is flushed on drop, and if flushing fails a panic
is thrown, which prints a crash message to the terminal. Since we can't
do anything if the write fails, and because file and stream writes are
approached with a best-effort attitude, we suppress the errors to
prevent the user from seeing them and getting concerned.
|
| |
|
|
|
|
| |
Writing to the input transmission port of the standard input stream no
longer clears the input queue. The only way to clear an input queue is
to write to the associated queue port.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The project previously used the `time` crate for getting the current
time in the local timezone. It was discovered that the time crate would
fail to return the local time when the program used multiple threads,
because of concerns that the other threads could modify timezone-related
environment variables while the timezone was being read from the system
and create a soundness issue. A second program thread was introduced
recently by commit 1a830a3 to provide non-blocking reads of standard
input for the local stream device, and so the time returned by the
clock device was falling back to UTC.
To fix this, the `time` crate has been replaced by the `chrono` crate
which does not suffer from this restriction, and in any case seems to
be the more popular of the two anyway.
|
| |
|
|
| |
This more accurately reflects its function.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is a complete rewrite and restructure of the entire emulator
project, as part of the effort in locking down the Bedrock specification
and in creating much better tooling for creating and using Bedrock
programs.
This commit adds a command-line argument scheme, an embedded assembler,
a headless emulator for use in non-graphical environments, deferred
window creation for programs that do not access the screen device,
and new versions of phosphor and bedrock-core. The new version of
phosphor supports multi-window programs, which will make it possible to
implement program forking in the system device later on, and the new
version of bedrock-core implements the final core specification.
|
| |
|
|
|
| |
The mapping of draw codes to draw operations didn't match the screen
device specification. This error was introduced in commit 6b3796c.
|
| |
|
|
|
|
|
|
| |
The most recently closed directory listing is cached instead of being
discarded. The next time a directory is to be opened, if the file path
is the same as the cached directory, the cached directory listing is
used in order to save the entire listing from having to be regenerated.
This is expected to result in performance gains on Windows.
|
| |
|
|
|
| |
The base variable is only used when compiling on Windows, so an
unused variable warning is raised when compiling on Linux.
|
| |
|
|
|
|
|
|
|
| |
Previously when reading from a ReadOnlyTextBuffer the pointer would
fall off the end and crash the emulator.
This commit also makes the ReadOnlyTextBuffer Unicode-compatible,
rather than the previous behaviour of clamping every character into
the ASCII range.
|
| |
|
|
|
|
|
| |
We were finding the parent of the relative path and then passing this
to the BedrockFilePath::from_path constructor, but this constructor
expects to be passed an absolute path and so was unconditionally
returning None.
|
| |
|
|
|
|
|
|
| |
This is the Windows side of the refactoring job. The windows crate has
been added as a dependency in order to get a list of available drives
by drive letter, and a virtual top-level root directory has been
implemented in the Windows code to make it possible for programs to
hierarchically navigate between available drives.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The major feature of this refactor is the creation of BedrockFilePath,
a type which can be losslessly converted into an operating-system
specific path or a Bedrock-style byte path. It also prevents file paths
which can't be represented as Bedrock-style byte paths from being
constructed, and allows the use of a base directory which acts as an
inescapable sandbox.
BedrockFilePath will support the creation of a virtual root directory
on Windows which will allow navigation between drive letters via the
standard hierarchical navigation ports.
This commit has been tested on Linux, but not yet Windows. Further work
is expected before the new code will work on Windows.
|
| |
|
|
| |
This should have been happening previously, but was missed.
|
| |
|
|
|
| |
Some Windows-only library functions were added in during a previous
commit while I was developing on Windows.
|
| |
|
|
|
|
|
|
| |
Construction of OsStrings is handled differently between Windows and
Unix, using platform-specific APIs.
The method OsStr::as_bytes was also changed to OsStr::as_encoded_bytes
at some point between Rust versions 1.69 and 1.80.
|
| |
|
|
|
|
|
| |
It was previously assumed that every entry in a directory would be
either a file or a directory, but a non-file and non-directory entry
has been encountered in the wild. In this case, just ignore the file,
leaving it out of the directory listing.
|
| |
|
|
|
| |
This is to allow the program to detect whether the previous ascend or
descend operation succeeded.
|
| | |
|
| |
|
|
|
|
| |
This is a temporary fix to make directories more navigable. This will
be removed when Cobalt can hide dot-prefixed filenames from directory
listings by itself.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Removed the sandbox directory of the file device, so the file device is
now sandboxed to the root directory. The sandboxing code has been left
intact for future use.
Permission flags have been added to the file device to allow each
category of file operation to be enabled or disabled in the emulator.
The default directory is set to be the current working directory from
when the emulator boots. This is the directory which is used when the
program attempts to ascend to the parent directory when no entry is
currently open.
|
| | |
|
| | |
|
| | |
|
| | |
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
| |
There was confusion between the role of the page_limit variable in the
memory device, where it was being used to mean both the absolute upper
limit of pages that are allowed to be allocated, but also to mean the
number of pages that the program would like to be allocated when needed.
To fix this, the program is now just told the full number of pages that
it is allowed to use, without first requiring that the program request
a particular number of pages.
|
| |
|
|
|
|
|
| |
The memory device was never being allocated any memory, the input device
wasn't scrolling correctly, the screen device was rendering sprites
incorrectly, and the device ID of the file device had not been changed
over from 0xA to 0x9.
|
| | |
|
| |
|
|
|
| |
This is a complete redesign of the file device. The most notable
addition is the ability to ascend and descend the file tree.
|
| |
|
|
|
| |
The stream device pointers now use wrapping semantics in order to
prevent overflow errors.
|
| |
|
|
|
|
|
|
| |
Wrapping stdin and stdout with BufReader and BufWriter makes reads and
writes more efficient by bundling multiple characters into a single
read or write operation. In practice, this hasn't made a difference to
running the numbers demo, but that demo spends the majority of processor
time converting values to decimal strings.
|
| |
|
|
|
| |
An effort to make the code shorter and more readable. This mostly
affects the code for the clock device.
|
| |
|
|
|
|
| |
The "Cumulative seconds" port of the CLOCK device has been changed
to be called "Cumulative timer", with the units changing from seconds
to 1/256 second ticks.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The vector line drawing operation of the screen device uses optimised
methods when drawing horizontal and vertical lines. The emulator now
interprets the coordinates of these lines as signed values, to allow for
the correct rendering of orthogonal lines where the coordinates of the
left or top edges are off-screen. Diagonal lines are yet to be tackled.
The bounding-box calculations shared by the rectangle and orthogonal
line drawing methods have also been split off into their own method,
significantly reducing the code duplication between all four of the
methods which require this functionality.
This commit also makes aesthetic changes to the code in the
draw_diagonal_line method to improve readability and brevity, as a part
of the changes made to the draw_line method when the points parameter
was removed from each of the three lower line-drawing methods.
|
| |
|
|
|
| |
This method was exceeding the 80-character line limit, which was
interfering with the centering behaviour of my text editor.
|
| |
|
|
|
| |
The method was between the two draw_sprite methods, but it fits better
just after draw_pixel.
|
| |
|
|
|
|
|
| |
The vector+size draw operations of the screen device draw textured and
untextured rectangles. The emulator now interprets the coordinates of
a rectangle as signed values, to allow for the correct rendering of
rectangles where the coordinates of the left or top edges are off-screen.
|
| |
|
|
|
|
|
| |
The vector+size+sprite draw operation of the screen device has been
changed in the specification from drawing a 1-bit textured triangle to
drawing a 1-bit textured line. This draw operation had not yet been
implemented, so this commit does not affect emulator functionality.
|
| |
|
|
|
|
| |
The order of the two result values from the widening multiply operation
in Rust was reversed, leading to incorrect multiplication results when
using the math device.
|
| |
|
