[html] tokenizer state holds attribute value and runs OOM

[TotzkePaul]

• VSCode Version: 1.3.0-insiders
• OS Version: Windows 10 Enterprise

Steps to Reproduce:

1. Install VS Code
2. Open Folder with 12 Mb html file.
3. "The window has crashed"

There isn't a noticeable increase in memory or CPU usage. Running as Admin or opening it as a single file makes no difference.

This seems similar to "Crashing/failing to open on new install #8400" except that it seems that I need to open a file first.

[bpasero]

@TotzkePaul does it reproduce with our latest insiders build (see below)? can you attach the file please?

We are releasing preview releases of the next stable VS Code version for everyone to try and give feedback. These preview releases are not 100% tested and might be unstable but contain our latest features and bugfixes. You can give our preview releases a try from: http://code.visualstudio.com/Download#insiders

[TotzkePaul]

It still fails on the latest nightly build.

It seems to be because there is a large amount of data in a value attribute. Removing that tag but leaving that data behind in a div doesn't cause an issue.

I was able to reproduce with non-useful data here:
https://github.com/TotzkePaul/SpacesVsTabs/blob/master/Test.html#L1

EDIT: I also have this issue with Atom. Maybe this isn't an issue with VS Code. Electron issue?

[alexdima]

The crash is due to OOM

[alexdima]

The problem is that the hand-written HTML tokenizer tries to build up the attributeValue in its State object and holds on to O(N^2) memory because each state is cloned at the beginning of a line.

[alexdima]

[TotzkePaul]

So I removed all the "\n\r" from the file and I got "The window is no longer responding. You can reopen or close the window or keep waiting. I'm not sure if I understood what you meant.

[jrieken]

@aeschli Correctly if I am wrong but we will change to textmate for html this milestone, right?

[aeschli]

@jrieken Not sure if already in this milestone.
As it's going to stay in the standalone editor anyway, I will start working on a fix.
From what I see it's not really necessary to have the attribute value as part of the tokenizer state.

[TotzkePaul]

I tested it out and works great! thanks!