Silent opt-out telemetry is harmful

[clason]

Lua-language-server has a telemetry function that is enabled by default.

While I understand that telemetry is useful and that making it opt-in will significantly limit that usefulness (because almost no-one will actively enable it), in times of increasing concern for privacy and data autonomy, this kind of silent telemetry (enabled by default without notification, no prominent mention in the README) will needlessly harm users' trust in the project. (To be clear, I'm not accusing you of bad faith here! I'm just describing how people react to such a feature in general.)

I therefore suggest the following: If the telemetry.enable config key is not set (to either true or false), lua-language server should print a message on startup saying

1. that the project uses telemetry,
2. where to find information about the collected data (what and where),
3. how to disable it (or enable it to remove the notice).

In addition, it might be better to put the whole block https://github.com/sumneko/lua-language-server/blob/6ba92a465c2686fe421fc398251d45c5d4381061/script/service/telemetry.lua#L62-L87 behind a if config.config.telemetry.enable guard so the timers aren't even run if the setting is not enabled.

[phmarek]

I vote for disabling telemetry be default; yes, you can print a message on first start, but enabling unless the user explicitly changes config files is the wrong way to go.

Yes, that may make the function useless - IMO it shouldn't be there at all!

[runiq]

I'd like to reiterate a few points from #463 here:

• Right now, the way telemetry handling is done is a GDPR violation
• A message at the end of the installation routine is probably a good idea

For everything else, I defer to the OP :)

Edit: To be even more clear, this will be a GDPR violation until it is made opt-in by default, as far as I understand the law.

[teto]

thanks for the great software but a bit mad at this being enabled by default. Care to explain the advantages of telemetry ? compared to bug reports let's say.

[sumneko]

While I understand that telemetry is useful and that making it opt-in will significantly limit that usefulness (because almost no-one will actively enable it), in times of increasing concern for privacy and data autonomy, this kind of silent telemetry (enabled by default without notification, no prominent mention in the README) will needlessly harm users' trust in the project. (To be clear, I'm not accusing you of bad faith here! I'm just describing how people react to such a feature in general.)

I will mention it in README. Currently it has been mentioned in changelog and setting

[sumneko]

I therefore suggest the following: If the telemetry.enable config key is not set (to either true or false), lua-language server should print a message on startup saying

1. that the project uses telemetry,
2. where to find information about the collected data (what and where),
3. how to disable it (or enable it to remove the notice).

I inspected some other software at the time. The telemetry function of VSCode is turned on by default. I just confirmed that he only mentioned this in the "I have read and agreed" of the giant during installation, and there has been no runtime prompt since then. .
Many other language services with telemetry capabilities do not even provide the option to turn off. I don’t know how you found out that I have telemetry capabilities to initiate a protest, I hope it’s not because of the shutdown option I kindly provided.

[sumneko]

In addition, it might be better to put the whole block

https://github.com/sumneko/lua-language-server/blob/6ba92a465c2686fe421fc398251d45c5d4381061/script/service/telemetry.lua#L62-L87

behind a if config.config.telemetry.enable guard so the timers aren't even run if the setting is not enabled.

This is an easy way to support immediate application of telemetry option changes without restarting the language service. I'm lazy, and I don't want to write so much code to support perfect timer management or pop up to notify users that they need to restart to take effect.

[runiq]

I inspected some other software at the time. The telemetry function of VSCode is turned on by default. I just confirmed that he only mentioned this in the "I have read and agreed" of the giant during installation, and there has been no runtime prompt since then.

I agree that's disingenious and par for the course for MS, but they do tell the user up front what they're doing, and that's still strictly better than just silently enabling things: If I read the TOS (I know, BIG if), and then decide I don't want to use the software because it phones home, I can do that.

That's also why I suggested to put a notice right at the end of the installation routine :)

[phmarek]

This is an easy way to support immediate application of telemetry option changes without restarting the language service. I'm lazy, and I don't want to write so much code to support perfect timer management or pop up to notify users that they need to restart to take effect.

But it also needlessly eats resources.

Please just take the telemetry out again, and provide a simple way to provide data - like the :HealthCheck function in nvim, for example.

[clason]

I inspected some other software at the time. The telemetry function of VSCode is turned on by default. I just confirmed that he only mentioned this in the "I have read and agreed" of the giant during installation, and there has been no runtime prompt since then.

The EULA does make a difference, legally -- and I would caution copying commercial projects with a big legal department that can handle any lawsuits arising from this. I need to repeat what @runiq mentioned: The way telemetry is implemented now is illegal in Europe (it violates the GDPR). While you of course can't be sued about it, it does make it illegal for me to use it in the context of my work.

Many other language services with telemetry capabilities do not even provide the option to turn off. I don’t know how you found out that I have telemetry capabilities to initiate a protest, I hope it’s not because of the shutdown option I kindly provided.

No, it was because a firewall flagged suspicious activity.

[sumneko]

I vote for disabling telemetry be default; yes, you can print a message on first start, but enabling unless the user explicitly changes config files is the wrong way to go.

Yes, that may make the function useless - IMO it shouldn't be there at all!

I can change it to a pop-up window to inform you of this during runtime, but I insist on enabling the telemetry function by default.
The telemetry function is very useful, and it means a lot to me.
Let me talk about the error reporting function first. I fixed about 30 runtime errors through the error log. You can find a lot of FIX a lot of runtime errors in the change log.
The reason for my initial development of this project is simple: I am a user of Lua, but VSCode does not have a relatively powerful Lua language service, so I decided to make one of my own and only use it for myself and my colleagues in the company.
As the project gradually iterated, I found that the project became the mainstream lua language service in VSCode. Now that many people are using it, I continue to develop and iterate, and this will last for 3 years. During these 3 years, I have not received any income (but a friend once said on Steam that he would give me a game, but the game type did not meet my preferences, so I refused). No one has participated in the long-term contribution (I used Chinese to submit at the beginning, but some people suggested that I change to English so that others can participate, and I did it).
The only thing that supports my continued development is that I think many people are using it. My company has at least 50 people using it, and there are many companies that use lua to develop games (now the mobile game is a development boom), so I assume there are 10,000 people using it. Until I counted the number of people online through the telemetry function, there were only 1,000 people!
I was really frustrated for a while, but I finally decided to continue to update and set a goal: 10,000 people use it. Currently you can find online statistics at http://119.45.194.183/. When I was writing this text, there were 3,600 people. That's why I say that he means a lot to me.

[clason]

I understand; that's why I didn't insist on removing telemetry or turning it off by default. It's the "silent" part I (and the GDPR) have problems with.

And let me say that it's an excellent server and your support is really amazing! (Otherwise I wouldn't have bothered with opening an issue and just stopped using it.)

[sumneko]

I'd like to reiterate a few points from #463 here:

• Right now, the way telemetry handling is done is a GDPR violation
• A message at the end of the installation routine is probably a good idea

For everything else, I defer to the OP :)

Edit: To be even more clear, this will be a GDPR violation until it is made opt-in by default, as far as I understand the law.

The installation process is implemented by the client, and I cannot insert instructions. What I can do is to tell you this through a pop-up notification every time you start the service, until you choose not to remind.

[clason]

The installation process is implemented by the client, and I cannot insert instructions. What I can do is to tell you this through a pop-up notification every time you start the service, until you choose not to remind.

That is exactly what I am proposing: If the telemetry.enable setting is not set by the user (either to true or false), send a notification. I'd be happy to help with finding a clear and concise wording for this notification.

[sumneko]

thanks for the great software but a bit mad at this being enabled by default. Care to explain the advantages of telemetry ? compared to bug reports let's say.

Although I am not willing to admit it, Lua is really not a mainstream programming language. At least in China, 80% of Lua users are game planners rather than programmers. They use lua to write data configurations, simple scripts, and damage calculation formulas. This means that they will not give any feedback when they encounter problems: "Github, what is that? How is it in plain English? Why do I have to register for an account to report problems?".