#579: Make content script injector aware of content script itself #661

fregante · 2021-06-30T15:04:52Z

Part of Injection improvements #579
Replaces #579: Deduplicate injection and make it more straightforward #654

To avoid sending huge PRs, I'm splitting this work over a few PRs. In this one I'm trying to preserve the general behavior while setting it up for further potential improvements. This PR:

Does away with repeated polling
Avoids duplicate injection (tabs with permanent permissions are just handled by webext-dynamic-content-script)
Creates a single point to "ready the content script", whether it's just waiting for the webext-dynamic-content-script, attempting to inject on a forbidden page, or manually injecting it

Ready except:

Pick better pbReady message
Pick better pbReady global

General direction for future PRs:

the lifters/messengers should just attempt to send the message first and then await/inject the content script automatically

We'll see how practical that will be though. The intent is avoid/limit calling the function itself and also to review the injection and communication overall.

src/contentScript/context.ts

src/background/util.ts

src/background/devtools/protocol.ts

fregante · 2021-06-30T15:34:33Z

src/background/devtools/internal.ts

      console.debug(
-        `Skipping injectDevtoolsContentScript because no activeTab permissions for tab: ${tabId}`
+        `Skipping injectDevtoolsContentScript because no activeTab permissions`,
+        { tabId, frameId, url }


The interesting part here is that activeTab is preserved for the top frame’s origin, but not for the iframe’s origins. This means that if a cross-origin iframe reloads, we lose access to it.

Tested on https://iframe-test-page.vercel.app/cross-origin-parent.html

You mean if the devtools are connected to another frame on the tab? (We haven't really though through the page editor functionality/handling for frames)

Please leave this note in the code comments, as it's important context for future work

I was talking about its behavior in general. I haven't really looked into how/why it deals with frames here other than to ensure that it throws errors correctly.

Will add the comment 👍 (Edit: added)

Ideally, probably in a future PR, I'll want to improve this part as well, like avoiding injection altogether if the tab never had activeTab in the first place, or if its URL was off-limits (chrome://)

src/background/util.ts

fregante · 2021-06-30T15:49:51Z

src/background/util.ts

+export async function onReadyNotification(signal: AbortSignal): Promise<true> {
+  return new Promise((resolve) => {
+    const onMessage = (message: unknown) => {
+      if (message === "pbReady") {


Likewise here I'm using a "raw" message because it's a temporary listener and MESSAGE_CONTENT_SCRIPT_READY is already fully handled in a different file. Let me know if you'd to follow the standard type: SOMETHING, payload: [] format instead and where I should put that constant.

For consistency, I have the preference for using {type: PIXIEBRIX_READY}. Even if we keep the string, I have the preference for using a shared constant for sending/handling the message

Maybe throw it in background/protocol for now? As part of the messaging overhaul in the future, we'll be figuring how where/how to refactor the action constants/payload typings

background/protocol has a lot of dependencies and code, I'll probably use messaging/constants.ts for now. Related:

Prevent context-specific code from being imported in the wrong context #669

src/contentScript/context.ts

twschiller · 2021-07-01T15:12:44Z

src/background/devtools/internal.ts

@@ -308,7 +323,7 @@ if (isBackgroundPage()) {
  });

  browser.webNavigation.onDOMContentLoaded.addListener((details) => {
-    void injectTemporaryAccess(details);
    emitDevtools("DOMContentLoaded", details);


Could you add a comment here on why this is the correct sequence of calls? It seems weird that we'd want the devtools to handle DOMContentLoaded before it has access

injectTemporaryAccess was not awaited, so I'd ask you the same question 😃
I moved it lower simply because the injector was logging before and after emitDevtools (because some code was execute before, and some after)

I think the order doesn't matter. As with everything involving communication, there's a lot of layers and it's hard to follow...

Eventually it gets to here: http://github.com/pixiebrix/pixiebrix-extension/blob/bc8c32d237056630d459b60bb445da5d373500b0/src/devTools/context.ts#L213-L213

Which calls http://github.com/pixiebrix/pixiebrix-extension/blob/bc8c32d237056630d459b60bb445da5d373500b0/src/devTools/context.ts#L132-L132

Which I think ends up handling the race condition through a combination of injectScript and waitReady

waitReady only exists for executeInTarget -> MESSAGE_RUN_BLOCK_TARGET now.

I haven't looked into emitDevtools("DOMContentLoaded") specifically, but if you expect it to run when the CS is ready, then it should await attemptTemporaryAccess.

src/contentScript/executor.ts

src/devTools/context.ts

twschiller · 2021-07-01T15:28:20Z

Overall this looks good. With complicated items like this, I have a strong preference for ensuring there's inline documentation of considerations, which sequencing is important, etc.

In general, reviewing this PR has affirmed that we need to have testing infrastructure in place prior to re-architecting the messaging infrastructure

Let me know when this is ready for testing and I'll help with manual QA. Please merge/rebase this PR, so we know what the impact is

fregante · 2021-07-01T17:29:21Z

Review followed. If you see 🚀 on your review messages it's because I followed them. I do this only when I also leave a comment, because if I "resolve" the conversation you might miss it.

I added some comments in the code, but in some places console.debug explained the situation well enough, I think. Let me know if you want me to expand on something else in the code.

fregante · 2021-07-01T17:30:42Z

package.json

@@ -220,7 +220,7 @@
      "^.+\\.txt$": "jest-raw-loader"
    },
    "transformIgnorePatterns": [
-      "node_modules/(?!@cfworker|webext-detect-page|idb|webext-patterns|webext-polyfill-kinda|p-timeout|p-defer)"
+      "node_modules/(?!@cfworker|webext-detect-page|idb|webext-patterns|webext-polyfill-kinda|webext-additional-permissions|p-timeout|p-defer)"


I'll look into enabling native ESM support in Jest at some point 😅 https://stackoverflow.com/a/61653104

twschiller

OK to merge - will address any connection corner cases as they arise

…self (#661)" This reverts commit 17739d6.

twschiller reviewed Jun 30, 2021

View reviewed changes

src/contentScript/context.ts Outdated Show resolved Hide resolved

twschiller reviewed Jun 30, 2021

View reviewed changes

src/background/util.ts Outdated Show resolved Hide resolved

twschiller reviewed Jun 30, 2021

View reviewed changes

src/background/util.ts Outdated Show resolved Hide resolved

Make content script injector aware of content script itself

482f559

fregante force-pushed the feature/injection branch from d3ec4a5 to 482f559 Compare June 30, 2021 16:06

Jest smh

ef6d152

fregante commented Jun 30, 2021

View reviewed changes

fregante marked this pull request as ready for review June 30, 2021 16:18

fregante marked this pull request as draft June 30, 2021 17:45

Let getTargetState errors bubble up + shorten logging

5398fdb

fregante marked this pull request as ready for review June 30, 2021 18:27

twschiller reviewed Jul 1, 2021

View reviewed changes

src/contentScript/executor.ts Outdated Show resolved Hide resolved

twschiller reviewed Jul 1, 2021

View reviewed changes

src/devTools/context.ts Show resolved Hide resolved

fregante added 3 commits July 2, 2021 00:02

Use symbol instead of window.pbReady

5958d9c

Use proper RPC format for messaging instead of raw string

b3917e2

(Nearby code changes)

92db848

fregante mentioned this pull request Jul 1, 2021

Prevent context-specific code from being imported in the wrong context #669

Closed

fregante added 4 commits July 2, 2021 00:08

Comment

2d80d97

Merge remote-tracking branch 'origin/main' into feature/injection

084c9f8

Simplify onReadyNotification with p-defer

f89b418

Comments 🎉

a88b64c

fregante commented Jul 1, 2021

View reviewed changes

twschiller approved these changes Jul 1, 2021

View reviewed changes

twschiller merged commit 17739d6 into main Jul 1, 2021

twschiller deleted the feature/injection branch July 1, 2021 19:53

twschiller added a commit that referenced this pull request Jul 1, 2021

Revert "#579: Make content script injector aware of content script it…

fcda39a

…self (#661)" This reverts commit 17739d6.

fregante mentioned this pull request Jul 2, 2021

Can't use page editor on page that doesn't have permissions #671

Closed

This was referenced Jul 4, 2021

Injection improvements #579

Closed

ensureContentScript does not inject styles.css #727

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

#579: Make content script injector aware of content script itself #661

#579: Make content script injector aware of content script itself #661

fregante commented Jun 30, 2021 •

edited

Loading

fregante Jun 30, 2021

twschiller Jul 1, 2021

fregante Jul 1, 2021 •

edited

Loading

fregante Jun 30, 2021

twschiller Jul 1, 2021

fregante Jul 1, 2021

twschiller Jul 1, 2021

fregante Jul 1, 2021

twschiller Jul 1, 2021

fregante Jul 1, 2021

twschiller commented Jul 1, 2021

fregante commented Jul 1, 2021 •

edited

Loading

fregante Jul 1, 2021

twschiller left a comment

#579: Make content script injector aware of content script itself #661

#579: Make content script injector aware of content script itself #661

Conversation

fregante commented Jun 30, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

fregante Jul 1, 2021 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

twschiller commented Jul 1, 2021

fregante commented Jul 1, 2021 • edited Loading

Choose a reason for hiding this comment

twschiller left a comment

Choose a reason for hiding this comment

fregante commented Jun 30, 2021 •

edited

Loading

fregante Jul 1, 2021 •

edited

Loading

fregante commented Jul 1, 2021 •

edited

Loading