Open Source Projects Europe

Thanks for the link. That helped me get going :)

I'll try to sketch the necessary things to make this work here (much for my own benefit).

webextension side

The webextension will just need to pass along messages to the native application side.

The code for sending messages is pretty simple:
javascript port = chrome.runtime.connectNative("<name>"); port.postMessage(message);

This will launch the application and keep the connection open.

Messages is encoded as json. Generating something like this in content.js passing it through background.js and then to the native application should work:

json { type: 'index', url: window.location.toString(), title: document.title, data: document.documentElement.outerHTML }

The webextension will also need an explicit addon id to be used by the native application manifest.

recoll application side

Instead of the script moving files from ~/Downloads we need a script which can receive messages. As luck would have it I already have a python script that can handle this (from an abandoned project using solr, solr is aweful...). The messaging functionality could be added to the recoll-we-move-files.py script, creating files in ~/.recollweb/ToIndex using the message content.

A native application manifest needs to be installed to let Firefox know about the native application.

The manifest will look something like this, where <name> is the name of the python script:

json { "name": "<name>", "description": "Example host for native messaging", "path": "/path/to/installed/python/script/<name>", "type": "stdio", "allowed_extensions": [ "<recollwe-addID>" ] }

On linux (been many years since I've used Windows so I'm not well versed on that) the manifest needs to be installed here /usr/lib/mozilla/native-messaging-hosts/<name>.json.

On windows a registry key needs to be set: HKEY_CURRENT_USER\SOFTWARE\Mozilla\NativeMessagingHosts\<name>.

The python script will need to parse the messages, look at the type (so it's possible to support search from the browser) and just save the data to an appropriately named file (along with the metadata), with no need for the intermediary file naming.

Search from the browser

With native messaging in place it should be possible to support searching from the browser, either from the address bar or from an extension popup. Address bar might be somewhat more convenient, but popups makes it easier to present snippets properly, though it's of course possible to support both setups.

Something like recoll -t -A -q <query>', where we get the query from a message, and then parsing the result into json sending it back ought to work OK, but would include matches that's not urls too, I'm guessing there's a way to filter based on source though. Filtering out anything that's not text/html might work okay too though.

Actually making it

In principle nothing needs to be included on the recoll side as all the native side needs to do is saving files in the correct directory. This makes prototyping easier. Ideally though it's best included in recoll so installation is just webextension + recoll, and not webextension + recoll + native app.

Hopefully I'll maintain the interest and put together a working prototype :)

Ah, yes, firefox will automatically launch the receiving script when the extension sends messages to it (by looking at the absolute path in the native application manifest).

rclbes:BGL is handy :)

I also realized that with a new web extension, recoll could just as easily support both extensions. That might be a cleaner approach, which wouldn't break anyone's existing workflows.

Thanks for taking a look :) As you noted it's far from ready yet.

For installation I think the ideal would be to bundle the native app with Recoll, so installation of the native app would be done by installing Recoll itself. This should work on most main stream distros I think. Unfortunately a stand alone installer which installs to $HOME seems to be needed for any distro that doesn't use FHS, eg. NixOS which I'm running.

I do intend to finish it and publish it to AMO, though haven't worked on it much lately. Will let you know when it's a more polished state.