Note that the ignore code for subprocess is being ported from the legacy worker watcher.
The Worker Target URL is now correctly set to the Absolute URL,
this has an impact on the context selector as we are displaying it.
Also, this patch highlights some issues around Common JS Loader and the distinct compartment
to be used when debugging privileged page, or when running the browser toolbox.
It caused leaks when running browser toolbox tests in debug builds.
The JS Process Actor is directly loaded in a distinct global and compartment "DevTools global",
thanks to `loadInDevToolsLoader` flag of BrowserToolboxDevToolsProcess JS Process Actor configuration.
There is no need to use `useDistinctSystemPrincipalLoader` helper as Loader.sys.mjs, loaded from
the JS Process Actor will also be loaded in the distinct global.
Then, we only need to ensure that the CommonJS Loader will also pick up the "DevTools global"
when it is loaded in that distinct global.
Things will be significantly easier once we migrate from CommonJS to ESM as the `{ global: "contextual" }`
option passed to `ChromeUtils.importESModule` will do the right thing, as well as static `import` statements.
Differential Revision: https://phabricator.services.mozilla.com/D204874
Note that the ignore code for subprocess is being ported from the legacy worker watcher.
Also, this patch highlights some issues around Common JS Loader and the distinct compartment
to be used when debugging privileged page, or when running the browser toolbox.
It caused leaks when running browser toolbox tests in debug builds.
The JS Process Actor is directly loaded in a distinct global and compartment "DevTools global",
thanks to `loadInDevToolsLoader` flag of BrowserToolboxDevToolsProcess JS Process Actor configuration.
There is no need to use `useDistinctSystemPrincipalLoader` helper as Loader.sys.mjs, loaded from
the JS Process Actor will also be loaded in the distinct global.
Then, we only need to ensure that the CommonJS Loader will also pick up the "DevTools global"
when it is loaded in that distinct global.
Things will be significantly easier once we migrate from CommonJS to ESM as the `{ global: "contextual" }`
option passed to `ChromeUtils.importESModule` will do the right thing, as well as static `import` statements.
Differential Revision: https://phabricator.services.mozilla.com/D204874
We seem to never wait for the breakpoint to be actually visibile before calling assertBreakpoint,
but it seems that it is relevant in this particular test, in the uncompressed codepath.
Differential Revision: https://phabricator.services.mozilla.com/D210991
We were waiting for Sources, which are resources, but not the targets,
which are updated with a different logic and timings.
Differential Revision: https://phabricator.services.mozilla.com/D210943
The previous patch explicitly prevents spawning JavaScriptTracer for same-process iframe,
as the JavaScriptTracer can only be spawn once per thread.
But then the JavaScriptTracer should ensure tracking DOM Events for these same-process iframe globals.
`DebuggerNotificationObserver` requires to connect to each individual WindowGlobal.
Connecting to the top WindowGlobal isn't enough to receive DOM Events for children WindowGlobals.
Differential Revision: https://phabricator.services.mozilla.com/D208468
* The formers JSWindow Actors used independently to create Target Actor for WindowGlobals, Web workers and Service workers
are all unified behind a single JSProcess Actor pair which was only used to create target actors for DOM Content Processes.
* The DevToolsProcess JSProcess actor now actively monitor the currently watched target types
in order to start and stop listeners specific to each target type.
We no longer rely on JSWindow Actors to observe WindowGlobal instantiations.
* `watchedByDevTools` is now consistantly set from each WindowGlobal's content process, via the WindowGlobalTargetActor.
With the new setup, the parent process no longer track WindowGlobal/BrowsingContext's and so there is no
natural place to flag them. While, in the content process, the target actor is an obvious place.
There is just one trick in window global target watcher in order to also set this flag on initial about blank documents.
* `browser-element-host` Session Data is now slightly more simple. It works like any other similar session data attribute.
But we have to ignore any pending exception coming up from updateDomainSessionDataForServiceWorkers call as that be still pending
while the toolbox closes, which make the JS Process Actor be unregistered,
and ultimately make the underlying sendQuery promise be rejected.
* The `loader` is made specific to each Watcher/connection and put in the connections map,
so that we can better support content toolbox & browser toolbox usecases
and we have a unique place to define the loader per watcher/connection.
* browser_target_command_detach.js: detach is no longer enough. It wouldn't stop watching and prevent the re-instantiation
of the target actor. We now have to stop watching for targets. This actually better match what happens on toolbox closing.
This test was meant as a unit test for covering remote debugging reusing the same DevToolsClient for debugger the same tab.
* `browser_toolbox_watchedByDevTools`. The management of watchedByDevTools is now driven from the content process.
And because the DevTools Server and especially the watcher actor is destroyed without an explicit RDP request
that the toolbox destruction is waiting for, we can't rely on closeToolbox to know
when the server is fully cleaned up.
Differential Revision: https://phabricator.services.mozilla.com/D203409
* The formers JSWindow Actors used independently to create Target Actor for WindowGlobals, Web workers and Service workers
are all unified behind a single JSProcess Actor pair which was only used to create target actors for DOM Content Processes.
* The DevToolsProcess JSProcess actor now actively monitor the currently watched target types
in order to start and stop listeners specific to each target type.
We no longer rely on JSWindow Actors to observe WindowGlobal instantiations.
* `watchedByDevTools` is now consistantly set from each WindowGlobal's content process, via the WindowGlobalTargetActor.
With the new setup, the parent process no longer track WindowGlobal/BrowsingContext's and so there is no
natural place to flag them. While, in the content process, the target actor is an obvious place.
There is just one trick in window global target watcher in order to also set this flag on initial about blank documents.
* `browser-element-host` Session Data is now slightly more simple. It works like any other similar session data attribute.
But we have to ignore any pending exception coming up from updateDomainSessionDataForServiceWorkers call as that be still pending
while the toolbox closes, which make the JS Process Actor be unregistered,
and ultimately make the underlying sendQuery promise be rejected.
* The `loader` is made specific to each Watcher/connection and put in the connections map,
so that we can better support content toolbox & browser toolbox usecases
and we have a unique place to define the loader per watcher/connection.
* browser_target_command_detach.js: detach is no longer enough. It wouldn't stop watching and prevent the re-instantiation
of the target actor. We now have to stop watching for targets. This actually better match what happens on toolbox closing.
This test was meant as a unit test for covering remote debugging reusing the same DevToolsClient for debugger the same tab.
* `browser_toolbox_watchedByDevTools`. The management of watchedByDevTools is now driven from the content process.
And because the DevTools Server and especially the watcher actor is destroyed without an explicit RDP request
that the toolbox destruction is waiting for, we can't rely on closeToolbox to know
when the server is fully cleaned up.
Differential Revision: https://phabricator.services.mozilla.com/D203409
Register them as soon as we open devtools, instead of waiting for debugger panel opening.
So that breakpoints can work even if we did not open debugger panel yet.
So far, only debugger statement would force the opening of the debugger.
This allows to remove the redundant code from debugger frontend.
Also clarify in this patch that thread-utils exports thread configurations.
Differential Revision: https://phabricator.services.mozilla.com/D205772
waitForSource wasn't fully reliable to retrieve the source object.
Now that it is, we can remove a few useless usages of findSource following it.
Differential Revision: https://phabricator.services.mozilla.com/D203447
This code was only used for workers targets, which weren't passing breakpoints
from the server side via "sessionData" concept.
Instead, we were still relying on the frontend to pass them and only after that
manually requesting the thread to resume its execution.
Now, there is still a notion of "attach" around the thread actor.
While all the targets are immediately instantiating their related thread actor,
the thread actor isn't active by default.
It becomes active as soon as:
* SOURCE or THREAD_STATE resource is watched
* breakpoints are set via session data (breakpoint list actor)
* thread configuration are set via session data (thread configuration actor)
I had to do one of these things in order to keep test from working.
In the production code, we do most of these things by default on the toolbox startup.
We could revise this behavior by:
* always attaching the thread actor, but it may start pausing unexpectedly in some tests.
* decouple sources from breakpoints within the thread actor, so that watching for sources doesn't necessarily "attach" the thread actor and activate breakpoints/debugger statements.
* have a new thread configuration to explicitly enable the thread actor pausing features.
I'm often using thread configuration to enable pausing in tests.
This required to improve xpcshell test helpers to be able to use Watcher/Commands.
About browser_target_command_bfcache.js, the DevTools code is now so quick that the
back/forward requests at done too quickly one after the other and fail actually moving to
the other document.
Differential Revision: https://phabricator.services.mozilla.com/D166321
Since the bug 1879447 enabled AccessibilityUtils to exclude intervening generics from tab-in-tablist accessibility checks, there are 3 tests in Debugger devtools that are now passing the a11y-checks, thus we could remove the `fail-if` notations from their test manifest.
Differential Revision: https://phabricator.services.mozilla.com/D204463
waitForSource wasn't fully reliable to retrieve the source object.
Now that it is, we can remove a few useless usages of findSource following it.
Differential Revision: https://phabricator.services.mozilla.com/D203447
This code was only used for workers targets, which weren't passing breakpoints
from the server side via "sessionData" concept.
Instead, we were still relying on the frontend to pass them and only after that
manually requesting the thread to resume its execution.
Now, there is still a notion of "attach" around the thread actor.
While all the targets are immediately instantiating their related thread actor,
the thread actor isn't active by default.
It becomes active as soon as:
* SOURCE or THREAD_STATE resource is watched
* breakpoints are set via session data (breakpoint list actor)
* thread configuration are set via session data (thread configuration actor)
I had to do one of these things in order to keep test from working.
In the production code, we do most of these things by default on the toolbox startup.
We could revise this behavior by:
* always attaching the thread actor, but it may start pausing unexpectedly in some tests.
* decouple sources from breakpoints within the thread actor, so that watching for sources doesn't necessarily "attach" the thread actor and activate breakpoints/debugger statements.
* have a new thread configuration to explicitly enable the thread actor pausing features.
I'm often using thread configuration to enable pausing in tests.
This required to improve xpcshell test helpers to be able to use Watcher/Commands.
About browser_target_command_bfcache.js, the DevTools code is now so quick that the
back/forward requests at done too quickly one after the other and fail actually moving to
the other document.
Differential Revision: https://phabricator.services.mozilla.com/D166321
This will prevent re-decoding the characters on each react update.
This will also allow other component to manipulate decoded characters.
Differential Revision: https://phabricator.services.mozilla.com/D202445
Also allow logging trace whenever any DOM Element get added, removed or modified.
For now this is exposed only via the console command via:
:tracer --dom-mutations (for all types of mutation)
:tracer --dom-mutations add,attributes,remove (coma separated list of mutation types you want to listen for)
Also revise the handling of "isWorker" so that all checks have the best possible check.
Differential Revision: https://phabricator.services.mozilla.com/D201976
This should help connect preferences to server's target configurations.
Once we get rid of thread's actor attach sequence, we may be able to also fold the getThreadOptions's preferences into this new system.
Differential Revision: https://phabricator.services.mozilla.com/D202277
`devtools.debugger.features.overlay` boolean preference will control this paused overlay.
(same name as an old preference which used to do that)
Differential Revision: https://phabricator.services.mozilla.com/D201697
This allows select text without having the highlighting going in the way.
It also prevents flashing the line when clicking around in the editor.
But we still update the location in the reducer in order to update the footer and outline panel
accordingly to the new cursor location.
Differential Revision: https://phabricator.services.mozilla.com/D200726
It allows
* toggling Source Map support entirely
* open the source map URL (when a bundle file is currently selected)
* open the mapped source (original or bundle)
* toggle the "open original source by default" setting
* show source map status (source map error, is it original or bundle file, is this a regular source?)
Differential Revision: https://phabricator.services.mozilla.com/D187577
It allows
* toggling Source Map support entirely
* open the source map URL (when a bundle file is currently selected)
* open the mapped source (original or bundle)
* toggle the "open original source by default" setting
* show source map status (source map error, is it original or bundle file, is this a regular source?)
Differential Revision: https://phabricator.services.mozilla.com/D187577
We should probably expose new RDP methods to prevent relying on console evaluations for this.
We would still have some potential issues if the evaluated method is about a JS symbol
whose name is the console command.
Differential Revision: https://phabricator.services.mozilla.com/D200167
Because the JavaScript tracer is currently running in each process/thread/target actor independently,
it is more complex to make it record across navigations as it may be spawn in a distinct process.
While it is fine for stdout and webconsole outputs (we could just spawn many concurrent tracers in each process),
this is more complex for the experimental "profiler" output.
For now, the profiler output automatically stops on target destruction and will open the profiler result.
By having an option to start recording on next page load, it prevents starting the tracer and prevent logging traces
for the current WindowGlobal. It should help focus on the new document.
For the profiler output, it prevents having the profiler to show up for the previous WindowGlobal.
Sideby tweak:
* stop passing logMethod and consider, like other option to be coming from the preferences.
* fix confusing state when debugging a page running in the same process.
Differential Revision: https://phabricator.services.mozilla.com/D196874
This simplifies toggling the Tracer on all active targets.
But this will be especially useful in the next changeset.
This allows to have two distinct level of enabling:
* the target configuration
* the actual start of tracing done by the tracer (on user interaction or next page load)
Doing this allows to distinguish tracer simple enabling,
when "trace on next user interaction" is enabled,
where we can display a badge on the tracer icon to significate it isn't tracing just yet.
And actual start of the tracer, when the first user interaction happens,
where we can remove the badge.
Differential Revision: https://phabricator.services.mozilla.com/D198961
Ensure updating the tracing state from the TracerCommand,
so that its internal state is correct whenever we start tracing
via the console or debugger.
Differential Revision: https://phabricator.services.mozilla.com/D196961