“…instead of unnecessarily burdening all the present plugins with unsolicited notifications (for most of them), it would be better if plugins that want something like this, actively request (registered somehow?) that info being sent only to them by the N++.”

Indeed, it (adding a new API) will be the better solution than keeping all events of SCN_MODIFIED notification ON for some plugins which don’t need the all these events, or even for users WITHOUT any plugin.

In the worst case, users get all the plugins which need ALL the events of notification SCN_MODIFIED and the performance will be equal as the old configuration (v8.7.5 and former versions), not worse.
But in the most cases, the performance will always be better than before.

At the risk of appearing argumentative… then why disable those events in the first place?

I remember e.g. this performance hint from the Scintilla author:

“Notepad++, unlike SciTE, leaves most notifications enabled so each replace operation will send 5 notifications [BeforeDelete, DeleteText, InsertCheck, BeforeInsert, InsertText]. Command events are also enabled which sends a WM_COMMAND:EN_CHANGE for each notification. These may not all be needed …”

And here is my similar thought:

“…instead of unnecessarily burdening all the present plugins with unsolicited notifications (for most of them), it would be better if plugins that want something like this, actively request (registered somehow?) that info being sent only to them by the N++.”

The PR2 which reverts the excluded SCN_MODIFIED events is here:
https://github.com/notepad-plus-plus/notepad-plus-plus/pull/16128

I have done the test with a huge XML file (400 MB) to compare between the PR to revert & the PR1 with new API NPPM_ADDSCNMODIFIEDFLAGS, by doing the following steps:

1. Open the huge XML and quit Notepad++ - so the huge XML file stays in the session.
2. Reopen Notepad++, make sure the 1st line is on the top. Immediatly hit Ctrl-PageDown. And use the timer to count the time to need to show the bottom text with syntax highlighting.

Here’s my result:

• PR2 with WRAP_ON: 39.72 - 39.72 sec to get the bottom with syntax highlighting.

• PR2 with WRAP_OFF: 38 - the bottom line is reached immediately, and it takes 38 seconds to have syntax highlighting.

• PR1 with WRAP_ON: 12.63 - 12.63 sec to get the bottom with syntax highlighting.

• PR1 with WRAP_OFF: 12.88 - the bottom line is reached immediately, and it takes 12.88 seconds to have syntax highlighting.

Conclusion, for the sake of performance, let’s use the PR1. And let plugins add the events they need for not overloading Notepad++ performance.

Set MODEVENTMASK_ON with SC_MOD_INSERTTEXT | SC_MOD_DELETETEXT only: Adding additional flags generates excessive and unnecessary event messages, degrading performance.

More events, less performance.
In absolute terms, it does make sense to me.

was the basis for changing the default event mask. I suspect this might have been a hasty remark, perhaps not considering that just because these events are unnecessary to Notepad++ itself does not mean that they are unnecessary to plugins. (The example given for overhead that the commenter claims could be skipped by suppressing unnecessary events is for an event, SC_MOD_CHANGEINDICATOR, that is included in the new defaults and won’t be disabled — so apparently it was necessary. In any case, if Notepad++ is doing unnecessary things in response to SCN_MODIFIED events, the sensible thing is to verify that they are unnecessary and remove that code. I suspect there isn’t a lot of code that was added for no reason at all… this logic seems suspect to me.)

It’s true while excluding unnecessary events for Notepad++, the needs of plugins were overlooked unintentionally.
But since now there’s a new API for plugins to add the events they want to listen to, from my perspective, it might be a better way - especially since the plugins in question are not installed by all users.

I’m questioning whether there is any real evidence that changing which SCN_MODIFIED events are normally enabled (completely separate from temporarily disabling all events during certain activities internal to Notepad++) gains anything. And if anything, enough to warrant the disruption for existing plugins and creation of a new NPPM message?

That’s a good point. I will test it from my side.

There are several factors come together to enhance large files’ syntax highlighting performance. Furthermore, it’s about only the performance of switching-in a large file.
So I think it’s OK if one of factors (Scintilla notifications) is turned ON by plugins. The consequence is not significative IMO.

At the risk of appearing argumentative… then why disable those events in the first place?

I cannot claim to have completely followed all the discussion behind the change to improve handling in large files, but it looks like the change to the SCN_MODIFIED event mask was based on this:

https://github.com/notepad-plus-plus/notepad-plus-plus/pull/15981#issuecomment-2563388005

I think the improvements to “the performance of switching-in a large file” were connected to turning off all SCN_MODIFIED events during loading. It looks like a note about restoring the environment after loading that said:

Set MODEVENTMASK_ON with SC_MOD_INSERTTEXT | SC_MOD_DELETETEXT only: Adding additional flags generates excessive and unnecessary event messages, degrading performance.

was the basis for changing the default event mask. I suspect this might have been a hasty remark, perhaps not considering that just because these events are unnecessary to Notepad++ itself does not mean that they are unnecessary to plugins. (The example given for overhead that the commenter claims could be skipped by suppressing unnecessary events is for an event, SC_MOD_CHANGEINDICATOR, that is included in the new defaults and won’t be disabled — so apparently it was necessary. In any case, if Notepad++ is doing unnecessary things in response to SCN_MODIFIED events, the sensible thing is to verify that they are unnecessary and remove that code. I suspect there isn’t a lot of code that was added for no reason at all… this logic seems suspect to me.)

I’m questioning whether there is any real evidence that changing which SCN_MODIFIED events are normally enabled (completely separate from temporarily disabling all events during certain activities internal to Notepad++) gains anything. And if anything, enough to warrant the disruption for existing plugins and creation of a new NPPM message?

though it could still break existing plugins if their authors are not active and able to keep up with changes.

I think it’s not reasonable to stop evolving a project like Notepad++ due to unmaintained plugins.

How will this work in connection with NPPN_GLOBALMODIFIED ? Will SCN_MODIFIED messages requested by NPPM_ADDSCINTILLANOTIFS still be suppressed during Replace All operations?

There’s no connection between NPPM_ADDSCINTILLANOTIFS & NPPN_GLOBALMODIFIED. They work separately.

If excluding those notifications really does improve Notepad++ performance for large files, anyone with Columns++ installed would sacrifice that gain, even on files for which they were not using elastic tabstops (which is more likely to be the case with large files) — even those who never use that feature at all. Other plugins might face similar circumstances.

There are several factors come together to enhance large files’ syntax highlighting performance. Furthermore, it’s about only the performance of switching-in a large file.
So I think it’s OK if one of factors (Scintilla notifications) is turned ON by plugins. The consequence is not significative IMO.

Please let me know if the PR works for you.

(The technical details might dissuade other users from posting other regressions in this discussion, and if the notification-discussion continues after someone does post another regression, their post may get lost in between posts about this issue.)

How will this work in connection with NPPN_GLOBALMODIFIED? Will SCN_MODIFIED messages requested by NPPM_ADDSCINTILLANOTIFS still be suppressed during Replace All operations?

What’s needed is a complelmentary set of *_GET_* and *_SET_* messages like Scintilla has for most properties.

Proper usage would be sending the *_GET_* message first, masking the return value for the desired flag(s), then restoring the flags by sending the *_SET_* message with the old value after the plugin is finished modifiying them.

The change is for enhancing large files’ syntax highlighting performance:
https://github.com/notepad-plus-plus/notepad-plus-plus/commit/de9ffd2ea8507d033f7f111d8b48762f7d3b9436

Since it’s for improving the performance & not all the plugins use such notifications, it won’t be reversed. Instead, I will add a new message NPPM_ADDSCINTILLANOTIFS for adding the notifications that plugins can add what they need, just after recieving NPPN_READY

What do you think?

I think it’s better than just removing functionality, though it could still break existing plugins if their authors are not active and able to keep up with changes. At least it would provide a quick way to restore compatibility when authors are available.

How will this work in connection with NPPN_GLOBALMODIFIED? Will SCN_MODIFIED messages requested by NPPM_ADDSCINTILLANOTIFS still be suppressed during Replace All operations?

Another thought has come to me, but the implications are unpleasant.

Columns++ only uses Scintilla notifications when implementing elastic tabstops, which can be enabled or disabled per document. By default, even after enabling it for one file, it is disabled when loading another file that is over 1000 KB or 5000 lines.

If excluding those notifications really does improve Notepad++ performance for large files, anyone with Columns++ installed would sacrifice that gain, even on files for which they were not using elastic tabstops (which is more likely to be the case with large files) — even those who never use that feature at all. Other plugins might face similar circumstances.

At present, I have no suggestion as to how that could be avoided in practice, though. For my plugin, since the loss of SC_MOD_BEFOREDELETE will cause degradation in responsiveness, rather than outright inaccurate behavior — and there may be another, more complex way to restore the responsiveness — I will have some design considerations to weigh with or without NPPM_ADDSCINTILLANOTIFS.

Thank you, that sounds good to me.

Notepad++'s code is biodegradable ;)

Nice to know, I already have enough hazardous waste with my rusty code :-D

ok, so just to be sure I understand this correctly, in my case I would call

::SendMessage(nppData._nppHandle, NPPM_ADDSCINTILLANOTIFS, 0, SC_MOD_BEFOREDELETE | SC_MOD_BEFOREINSERT);
to reactivate these notifications?
And does this mean that these notifications are then globally available again?

Yes.

That is, if NppLspClient reactivates them, can PythonScript use them again?

Yes. But one plugin should not depend on other plugins’ registration.

Or are they then only forwarded to the respective plugin?

No.

By default, MODEVENTMASK_ON will contain the value SC_MOD_DELETETEXT | SC_MOD_INSERTTEXT | SC_PERFORMED_UNDO | SC_PERFORMED_REDO | SC_MOD_CHANGEINDICATOR

In the case that both Columns++ & NppLspClient are installed and both plugins have integrated NPPM_ADDSCINTILLANOTIFS message, and NppLspClient is loaded after Columns++:

After Columns++'s registration, MODEVENTMASK_ON will contain the value SC_MOD_DELETETEXT | SC_MOD_INSERTTEXT | SC_PERFORMED_UNDO | SC_PERFORMED_REDO | SC_MOD_CHANGEINDICATOR | SC_MOD_DELETETEXT | SC_MOD_INSERTTEXT.

But after NppLspClient’s registration, the value becomes SC_MOD_DELETETEXT | SC_MOD_INSERTTEXT | SC_PERFORMED_UNDO | SC_PERFORMED_REDO | SC_MOD_CHANGEINDICATOR | SC_MOD_DELETETEXT | SC_MOD_INSERTTEXT | SC_MOD_BEFOREDELETE | SC_MOD_BEFOREINSERT.

And the latest registered value will be sent globally for all installed plugins.

Btw, fun fact, isn’t there a law in France against deliberate obsolescence :-D

I ensure you that it’s not deliberate obsolescence.
Notepad++'s code is biodegradable ;)

ok, so just to be sure I understand this correctly, in my case I would call

::SendMessage(nppData._nppHandle, NPPM_ADDSCINTILLANOTIFS, 0, SC_MOD_BEFOREDELETE | SC_MOD_BEFOREINSERT);

to reactivate these notifications?
And does this mean that these notifications are then globally available again? That is, if NppLspClient reactivates them, can PythonScript use them again? Or are they then only forwarded to the respective plugin?

@donho said in Notepad++ v8.7.6 released:

NPPM_ADDSCINTILLANOTIF

Does this mean that a plugin registers with this new message when starting for certain notifications?

Yes

NPPM_ADDSCINTILLANOTIF

Does this mean that a plugin registers with this new message when starting for certain notifications?

The change is for enhancing large files’ syntax highlighting performance:
https://github.com/notepad-plus-plus/notepad-plus-plus/commit/de9ffd2ea8507d033f7f111d8b48762f7d3b9436

Since it’s for improving the performance & not all the plugins use such notifications, it won’t be reversed. Instead, I will add a new message NPPM_ADDSCINTILLANOTIFS for adding the notifications that plugins can add what they need, just after recieving NPPN_READY, for the next release:

extern "C" __declspec(dllexport) void beNotified(SCNotification *notifyCode)
{
	switch (notifyCode->nmhdr.code) 
	{

		case NPPN_READY:
		{
			::SendMessage(nppData._nppHandle, NPPM_ADDSCINTILLANOTIFS, 0, SC_MOD_DELETETEXT | SC_MOD_INSERTTEXT);
		}
		break;
...

What do you think?

Possibly introducing a subscription model would be performance relevant, but that would also mean that currently running but no longer maintained plugins would then be obsolete. (which is also true when removing plugin related functionality)

Btw, fun fact, isn’t there a law in France against deliberate obsolescence :-D

Yes, I think. Does your plugin need these notifications?

Yes, both messages ensure that I get the correct position data that I need to calculate and convert.
I’m not saying it can’t be done with SC_MOD_DELETETEXT and SC_MOD_INSERTTEXT, but it’s definitely more difficult to get it right, at least in my case.

EDIT: By the way, this also applies to script plugins such as PythonScript etc., as these normally export all available functions.

So I guess the question is: assuming NPPM_SCINOTIFICATIONENABLED always accurately reflected what messages are broadcasted, do you see it being enough of a benefit to you and other similarly situated people that it would be worth adding?

No.

I can really only speak for my own plugin. I had to figure out by trial and error what I could count on, because documentation isn’t all that precise. Experimenting with what I hope were realistic cases, I tried to figure out how I could maintain the desired display without lag. I found a way that works often enough. I chose a trade-off between complexity and results. I could make the implementation of elastic tabstops in Columns++ more responsive in more cases, but it wouldn’t improve the worst case, the typical case is already pretty good, and it would add a lot of complexity to the code and (in my estimation) make it more error-prone.

That trade-off depends on which messages I get. If I won’t be getting a notification before delete, I have to rethink that trade-off. I’ll probably conclude that I have to add the complex, error-prone code I’ve not yet written, because a rather common case will become laggy without it. It will take time, because it is not simple.

It wouldn’t be any help to find out dynamically what messages I will get. I need to know at design time what I can count on.

Would it be possible and practical for Notepad++ to have a way for plugins to tell it, at startup, what Scintilla notifications they need, so Notepad++ can avoid disabling them?

This makes me wonder if maybe Notepad++ could do the reverse: have an NPPM_SCINOTIFICATIONENABLED(uint sciNotificationValue) message that plugins could send to NPP, where NPP would return 1 if sciNotificationValue is broadcasted to plugins, 0 if it is not broadcasted, and -1 if it is not a valid Scintilla Notification value.

The main downside I see of this hypothetical message is that Don Ho and others are somewhat likely to forget to update the function, and then it becomes worse than useless.

So I guess the question is: assuming NPPM_SCINOTIFICATIONENABLED always accurately reflected what messages are broadcasted, do you see it being enough of a benefit to you and other similarly situated people that it would be worth adding?

@Ekopalypse said in Notepad++ v8.7.6 released:

8.7.6 no longer forwards SC_MOD_BEFOREDELETE and SC_MOD_BEFOREINSERT. Will it stay that way?

Yes, I think. Does your plugin need these notifications?

Would it be possible and practical for Notepad++ to have a way for plugins to tell it, at startup, what Scintilla notifications they need, so Notepad++ can avoid disabling them? I don’t really know enough about why you sometimes choose to disable notifications we used to get, so I don’t claim to know if this idea makes any sense.

no longer forwards SC_MOD_BEFOREDELETE and SC_MOD_BEFOREINSERT

Maybe when the Release Candidate is published, such removals should be listed in a “beware plugin authors” (or some such) section.

This could help for things that the author would be willing to revert (as he did the layout cache change). Perhaps (and I hope) he will revert this as well. But by the time a release candidate is finalized, it’s already too late to update plugins and get new versions into the plugin list, so it wouldn’t help with changes that will not be reverted.

We really need notice when potentially breaking changes become pull requests, so we can test and comment before they are accepted if we think they might adversely affect our plugins (and scramble to mitigate them if they are accepted anyway). I’m not sure how practical that is, though… it might be difficult for @donho to recognize a potentially breaking change. The structure of the Notepad++ plugin facility makes plugins potentially very powerful, but also potentially fragile.The Notepad++ messages are documented, but there are no documented “promises” (or boundaries) as to much of anything else, so it’s probably hard to tell what even constitutes a potentially breaking change.

no longer forwards SC_MOD_BEFOREDELETE and SC_MOD_BEFOREINSERT

Maybe when the Release Candidate is published, such removals should be listed in a “beware plugin authors” (or some such) section.