putImageData works a little differently to drawImage which confused me
for a bit, you need to offset the destination as putImageData draws the
'dirty' parts of the texture using the same source image dimensions
Fixes#4197
A common mistake made was overriding dispose and not calling super.dispose,
so a TS trick is now used to move it to a readonly property instead of a
member method which prevents it being overridden.
This is the proper fix for microsoft/vscode#162996, after a bunch of
investigation we found that this is causing the problem:
VS Code clears the texture atlas on OS resume to fix an issue where the texture
could be corrupted. This calls clearTextureAtlas twice which should be fine, but
it's not because inside WebglRenderer.clear we're calling updateModel which is
meant to only ever be called in an animation frame. This calls potentially both
renderers to have stale views of the texture atlas for those rows and our perf
caching of cells (the main purpose of the model) doesn't know it's stale.
Co-Authored-By: Megan Rogge <merogge@microsoft.com>
Add support for the parent element (what is passed to Terminal.open)
being in a different (same origin) window. Accesses to DOM APIs such
as requestAnimationFrame and devicePixelRatio thus need to be scoped
to the corrent window, instead of assuming that it's the same as the
window/global scope where the code is running.
This is done by inferring a parent window at the creation time, and
then storing it in CoreBrowserService, which is already passed to most
places that need it.
To catch future regressions, an ESLint rule that checks for global
accesses is added (it uses AST selectors via the no-restricted-syntax
rule).
This should also be applicable when the parent element is an iframe.
Fixes#3758
Daniel Imms
Sebastian Malton
coderaiser
Mihai Parparita
Jean Pierre