Use more universal ANSI sequence for 'clear screen and clear buffer' #57701

colinta · 2024-03-08T20:06:34Z

Fixes #57700
Fixes #29829
Fixes #57894
Closes #59315

Tiny tiny patch to bring just a little more terminal support to tsc --watch. In my case, iTerm2 needs this ESC [ 3J sequence in order to clear screen and remove scroll history.

If there's a setting in iTerm2 to change this, I would happily do that, but I didn't see anything.

  92049 passing (6m)

Finished do-runtests-parallel in 6m 22.9s
Completed runtests-parallel in 6m 57.3s
✨  Done in 417.81s.

mrazauskas · 2024-03-16T19:45:06Z

For what it’s worth, the ansi-escapes library has two similarly named methods: clearScreen() and clearTerminal().

clearScreen() is the same as current tsc --watch implementation.

clearTerminal() is more complex and it is based on Jest's implementation (see sindresorhus/ansi-escapes#6 and https://github.com/jestjs/jest/blob/6b120fbc5d1cadb3ef205bd1d48549155bed5dd1/packages/jest-util/src/specialChars.ts#L18-L20)

I am pointing to ansi-escapes, because it is rather widely used. Not sure if there are any standards in this field.

jakebailey · 2024-03-20T23:42:56Z

If we're going to do this, I'm somewhat inclined to take the more popular approach of using the same sequence that ansi-escapes/jest take, though that appears to be platform specific which may make our tests unhappy.

I'll have to ask for feedback on this from the team.

colinta · 2024-03-22T14:15:32Z

I wonder if we even need to make a distinction between mac and windows

echo -n -e "\x1B[2J\x1B[3J\x1BH\x1B[0f"

Works on all my terminals (Terminal, iTerm, & VSCode). I don't have a Windows machine to test on, though. (I updated the PR to use this sequence)

colinta · 2024-04-25T16:36:48Z

The passage of time made me forget the ANSI sequence I had settled on, but I remembered this morning and updated this PR one more time.

So, the sequence in there is based on the one from ansi-sequences. It's just a combination of the two, without the OS check.

jakebailey · 2024-06-14T23:40:34Z

I believe that this also fixes #29829 (which I found via people commenting duplicates on the related issues to this PR's).

Planning on doing some local testing, though, it's just a lot of terminal emulators to test...

If we are doing this, I hope nobody relies on our specific escape sequence. I think not, but rather they depend on other watch messages.

jakebailey · 2024-07-22T18:51:11Z

I did some testing of this PR and #59315.

old behavior:

linux, wsl vscode: can't scroll up ✅
linux, wsl terminal: can't scroll up ✅
linux, foot: can't scroll up ✅
linux, tilix (vte): can't scroll up ✅
windows 11, vscode, pwsh: can't scroll up ✅
windows 11, terminal, pwsh: can't scroll up ✅
windows 11, terminal, cmd: can't scroll up ✅
windows 11, console host, cmd: can't scroll up ✅
macos, kitty: can't scroll up ✅
macos, terminal: can scroll up ❌
macos, iterm: can scroll up ❌

We emit 0x1Bc, causing #57894 (and I assume #29829 too).

#57701

linux, wsl vscode: can't scroll up ✅
linux, wsl terminal: can't scroll up ✅
linux, foot: can't scroll up ✅
linux, tilix (vte): can't scroll up ✅
windows 11, vscode, pwsh: can't scroll up ✅
windows 11, terminal, pwsh: can't scroll up ✅
windows 11, terminal, cmd: can't scroll up ✅
windows 11, console host, cmd: can't scroll up ✅
macos, kitty: can't scroll up ✅
macos, terminal: (changed) can't scroll up ✅
macos, iterm: (changed) iterm asks for permission to clear screen, then can't scroll up ✅

No 0x1Bc, so should fix #57894.

#59315

linux, wsl vscode: can't scroll up ✅
linux, wsl terminal: can't scroll up ✅
linux, foot: can't scroll up ✅
linux, tilix (vte): can't scroll up ✅
windows 11, vscode, pwsh: can't scroll up ✅
windows 11, terminal, pwsh: can scroll up ❌
windows 11, terminal, cmd: can scroll up ❌
windows 11, console host, cmd: can scroll up ❌
macos, kitty: can't scroll up ✅
macos, terminal: can't scroll up ✅
macos, terminal: (changed) can't scroll up ✅
macos, iterm: (changed) iterm asks for permission to clear screen, then can't scroll up ✅

No 0x1Bc, so should fix #57894.

I have not tested on something like Windows 10 with cmd, though; but I'm fairly confident that this PR is good.

jakebailey · 2024-07-22T18:56:50Z

Using this launch task in our repo:

        {
            "label": "tsc: watch ./src with built",
            "type": "shell",
            "command": "node",
            "args": ["${workspaceFolder}/built/local/tsc.js", "--build", "${workspaceFolder}/src", "--watch"],
            "group": "build",
            "isBackground": true,
            "problemMatcher": [
                "$tsc-watch"
            ]
        },

Shows that the problem matcher in VS Code also works fine:

Then undo the bad code:

mrazauskas · 2024-07-22T19:04:35Z

Out of curiosity: what if \u001B[2J\u001B[3J\u001B[H is always emitted regardless of platform? Seemed like it is not covered in #57701 (comment). Or I misunderstood?

jakebailey · 2024-07-22T19:05:40Z

That's exactly what this PR does, if I'm understanding you correctly.

diff --git a/src/compiler/sys.ts b/src/compiler/sys.ts
index 32d7af09c03e7..cf9318afdd199 100644
--- a/src/compiler/sys.ts
+++ b/src/compiler/sys.ts
@@ -1619,7 +1619,7 @@ export let sys: System = (() => {
             setTimeout,
             clearTimeout,
             clearScreen: () => {
-                process.stdout.write("\x1Bc");
+                process.stdout.write("\x1B[2J\x1B[3J\x1BH\x1B[0f");
             },
             setBlocking: () => {
                 const handle = (process.stdout as any)?._handle as { setBlocking?: (value: boolean) => void; };

mrazauskas · 2024-07-22T19:07:02Z

That's exactly what this PR does, if I'm understanding you correctly.

Almost. I am suggesting to skip \x1B[0f.

jakebailey · 2024-07-22T19:08:51Z

I can test that, but what effect are you trying to get by omitting that?

jakebailey · 2024-07-22T19:12:43Z

Tested, and omitting that seems to leave a bunch of space above the log, not actually clearing the screen (and the extra empty lines are scrollable). So it seems to be required.

mrazauskas · 2024-07-22T19:15:19Z

Thanks. That’s interesting. I thought that \x1B[H and \x1B[0f both do the same thing. Only testing can proof that.

mrazauskas · 2024-07-22T19:21:46Z

Sorry, me again. Seems like \x1BH is missing [ in this PR, but has it in #59315. Shouldn’t it read \x1B[H?

ehoogeveen-medweb · 2024-07-22T19:21:50Z

If I'm reading this correctly, \x1BH sets a horizontal tab stop whereas \x1B[H is interpreted as pressing the Home key and \x1B[0f moves the cursor to a particular position in the viewport (it seems to take 2 coordinates so I'm not sure what only providing 1 implies).

Edit: Ah, the rules for the omission are noted below: "If <x> and <y> are omitted, they will be set to 1;1." So I guess \x1B[0f is equivalent to \x1B[0;1f?

Edit 2: I think I was misreading the bit about the Home key. It's the other way around - pressing the Home key in "Normal Mode" generates the CUP sequence \x1B[H which is equivalent to \x1B[1;1H... I think.

Edit 3: And going full circle, "If <n> is omitted or equals 0, it will be treated as a 1" so \x1B[0f == \x1B[0;1f == \x1B[1;1f and \x1B[H == \x1B[1;1H which should both do the same thing. But \x1BH (missing the [) should mean something else.

ehoogeveen-medweb · 2024-07-22T19:45:47Z

To add to the above, \x1B[3J should do nothing according to this as 3 is not a valid value for <n>. So technically the following should both be equivalent and sufficient:

\x1B[2J\x1B[H
\x1B[2J\x1B[f

But maybe there are differences between terminals that mean you need more.

jakebailey · 2024-07-22T19:49:42Z

I'm going through all of the combos now (fixing the potential typo, trying those two shorter sequences).

ehoogeveen-medweb · 2024-07-22T19:50:08Z

Ah, that page is incomplete - 3J indeed clears the scrollback, as commented here (and probably in other places). So you do need both:

\x1B[2J\x1B[3J\x1B[H
\x1B[2J\x1B[3J\x1B[f

ehoogeveen-medweb · 2024-07-22T20:02:51Z

According to this, CUP (\x1B[H here) should be preferred over HVP (\x1B[f here) for "new applications". Either way they should be equivalent :)

jakebailey · 2024-07-22T20:03:19Z

#57701 but with \x1B[H typo fix

linux, wsl vscode: can't scroll up ✅
linux, wsl terminal: can't scroll up ✅
linux, foot: can't scroll up ✅
linux, tilix (vte): can't scroll up ✅
windows 11, vscode, pwsh: can't scroll up ✅
windows 11, terminal, pwsh: can't scroll up ✅
windows 11, terminal, cmd: can't scroll up ✅
windows 11, console host, cmd: can't scroll up ✅
macos, kitty: can't scroll up ✅
macos, terminal: can't scroll up ✅
macos, iterm: can't scroll up ✅

Just \x1B[2J\x1B[3J\x1B[H

linux, wsl vscode: can't scroll up ✅
linux, wsl terminal: can't scroll up ✅
linux, foot: can't scroll up ✅
linux, tilix (vte): can't scroll up ✅
windows 11, vscode, pwsh: can't scroll up ✅
windows 11, terminal, pwsh: can't scroll up ✅
windows 11, terminal, cmd: can't scroll up ✅
windows 11, console host, cmd: can't scroll up ✅
macos, kitty: can't scroll up ✅
macos, terminal: can't scroll up ✅
macos, iterm: can't scroll up ✅

Just \x1B[2J\x1B[3J\x1B[f

linux, wsl vscode: can't scroll up ✅
linux, wsl terminal: can't scroll up ✅
linux, foot: can't scroll up ✅
linux, tilix (vte): can't scroll up ✅
windows 11, vscode, pwsh: can't scroll up ✅
windows 11, terminal, pwsh: can't scroll up ✅
windows 11, terminal, cmd: can't scroll up ✅
windows 11, console host, cmd: can't scroll up ✅
macos, kitty: can't scroll up ✅
macos, terminal: can't scroll up ✅
macos, iterm: can't scroll up ✅

However, not including H or f causes extra empty lines to be printed, so one of those must be present.

jakebailey · 2024-07-22T20:04:35Z

According to this, CUP (\x1B[H here) should be preferred over HVP (\x1B[f here) for "new applications". Either way they should be equivalent :)

That was the next question, which one to choose! Seems like we should be good with \x1B[2J\x1B[3J\x1B[H. Likely, the current PR only works because the typo didn't matter due to [f.

ehoogeveen-medweb · 2024-07-22T20:36:31Z

For completeness:

The old sequence \x1Bc (ESC c) was RIS (Reset to Initial State)
The new sequence \x1B[2J \x1B[3J \x1B[H (CSI 2 J CSI 3 J CSI H) is two ED (Erase in Display) sequences followed by a CUP (CUrsor Position) sequence
Of these, CSI 3 J to clear scrollback is something that wasn't supported on the VT510 but was added as an extension to xterm and adopted by other terminals; see e.g. here (Erase Saved Lines)

sandersn requested review from jakebailey and sheetalkamat March 20, 2024 13:33

sandersn assigned jack-williams and jakebailey and unassigned jack-williams Mar 20, 2024

sandersn added the For Backlog Bug PRs that fix a backlog bug label Mar 20, 2024

colinta force-pushed the main branch from 929f24d to cb948f5 Compare April 24, 2024 20:37

colinta changed the title ~~Use 'standard' ANSI sequence for 'clear screen and clear buffer'~~ Use more universal ANSI sequence for 'clear screen and clear buffer' Apr 24, 2024

use more universal ANSI sequence for 'clear screen and clear buffer'

f77bc14

colinta force-pushed the main branch from cb948f5 to f77bc14 Compare April 25, 2024 16:36

jakebailey mentioned this pull request Jul 17, 2024

Fix: tsc -w is losing working directory in macOS Terminal #59315

Closed

jakebailey approved these changes Jul 22, 2024

View reviewed changes

Simplify ansi escapes

44ac7fd

sheetalkamat approved these changes Jul 22, 2024

View reviewed changes

jakebailey merged commit 3f2a9e4 into microsoft:main Jul 22, 2024
32 checks passed

jakebailey mentioned this pull request Aug 26, 2024

Use more compatible screen clearing ANSI escape oven-sh/bun#13553

Merged

8 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Use more universal ANSI sequence for 'clear screen and clear buffer' #57701

Use more universal ANSI sequence for 'clear screen and clear buffer' #57701

colinta commented Mar 8, 2024 •

edited by jakebailey

Loading

mrazauskas commented Mar 16, 2024

jakebailey commented Mar 20, 2024 •

edited

Loading

colinta commented Mar 22, 2024 •

edited

Loading

colinta commented Apr 25, 2024

jakebailey commented Jun 14, 2024

jakebailey commented Jul 22, 2024 •

edited

Loading

jakebailey commented Jul 22, 2024

mrazauskas commented Jul 22, 2024

jakebailey commented Jul 22, 2024

mrazauskas commented Jul 22, 2024

jakebailey commented Jul 22, 2024

jakebailey commented Jul 22, 2024 •

edited

Loading

mrazauskas commented Jul 22, 2024 •

edited

Loading

mrazauskas commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024 •

edited

Loading

ehoogeveen-medweb commented Jul 22, 2024

jakebailey commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024

jakebailey commented Jul 22, 2024 •

edited

Loading

jakebailey commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024 •

edited

Loading

Use more universal ANSI sequence for 'clear screen and clear buffer' #57701

Use more universal ANSI sequence for 'clear screen and clear buffer' #57701

Conversation

colinta commented Mar 8, 2024 • edited by jakebailey Loading

mrazauskas commented Mar 16, 2024

jakebailey commented Mar 20, 2024 • edited Loading

colinta commented Mar 22, 2024 • edited Loading

colinta commented Apr 25, 2024

jakebailey commented Jun 14, 2024

jakebailey commented Jul 22, 2024 • edited Loading

jakebailey commented Jul 22, 2024

mrazauskas commented Jul 22, 2024

jakebailey commented Jul 22, 2024

mrazauskas commented Jul 22, 2024

jakebailey commented Jul 22, 2024

jakebailey commented Jul 22, 2024 • edited Loading

mrazauskas commented Jul 22, 2024 • edited Loading

mrazauskas commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024 • edited Loading

ehoogeveen-medweb commented Jul 22, 2024

jakebailey commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024

jakebailey commented Jul 22, 2024 • edited Loading

jakebailey commented Jul 22, 2024

ehoogeveen-medweb commented Jul 22, 2024 • edited Loading

colinta commented Mar 8, 2024 •

edited by jakebailey

Loading

jakebailey commented Mar 20, 2024 •

edited

Loading

colinta commented Mar 22, 2024 •

edited

Loading

jakebailey commented Jul 22, 2024 •

edited

Loading

jakebailey commented Jul 22, 2024 •

edited

Loading

mrazauskas commented Jul 22, 2024 •

edited

Loading

ehoogeveen-medweb commented Jul 22, 2024 •

edited

Loading

jakebailey commented Jul 22, 2024 •

edited

Loading

ehoogeveen-medweb commented Jul 22, 2024 •

edited

Loading