Zero width format

[mathstuf]

I decided to not do %0{ and %} because the logic for control sequences more than just a single character is a messier. I can also change it over to %{ and %} if separate open/close sequences are preferable. It is not (currently) an error to have mismatched open/close sequences, but it is also not a zero-width section and %0 will be literal in that case.

---

Fixes #713.

[mathstuf]

Ping? @martine? @nico?

[evmar]

I'm not so familiar with this corner of terminal IO. What is the normal syntax for zero-width markers in other similar languages? (bash/zsh?)

[mathstuf]

zsh uses %0{ and %} which is messy due to the two-character escape. bash seems to use \[ and \].

[nico]

The motivation here is to let eliding be correct if NINJA_STATUS contains ansi codes, right? Maybe we could have the eliding function not count ansi codes when computing widths?

[mathstuf]

I could for see using it to change tmux's title using its escape codes, but right now I'd be fine with color. Does the color stripping understand 256color sequences?

[nico]

Are they special? Currently it does this: https://github.com/martine/ninja/blob/master/src/util.cc#L477

[mathstuf]

Here's the command which does the work of setting the title for me:

printf "\033]2;%s\033\\" "$title"

[nico]

I meant the 256color sequences.

(Do I understand you correctly that you want your ninja status to modify your tmux title? Why?)

[mathstuf]

Ah, sorry, missed the context. Well, I don't personally, but this would also support that seamlessly (possibly counting down the targets?). Here's my 256-color code:

for color in {0..255}; do
    ZSH_COLOR[$color]=$'\033[38;5;'$color'm'
    ZSH_COLOR_PS1[$color]="%0{${ZSH_COLOR[$color]}%}"
done

It looks like the existing code should work.

[evmar]

@nico modifying your terminal title is commonly a hack for long-running processes to display status in a more visible place like alt-tab lists. Some WMs even let you "shade" a window (collapsing it to its title bar) as a substitute for minimization.

[nico]

Right, but is this something one would want to do every time a ninja command updates? (I honestly don't know.)

If we end up doing #486 (which I currently do want to do, to get colors from clang diagnostics on Windows), ninja will have some knowledge about ansi codes anyhow, so it doesn't seem unreasonable for Elide() to learn about them too maybe.

Are there other use cases for zero-width markers other than ansi codes?

[evmar]

From a pedantic perspective it's hard to know in advance the full
possibility space of escapes, see e.g.
http://tldp.org/HOWTO/Xterm-Title-6.html

(PS: there was actually a vulnerability in one of these in the terminal
escape parser -- it allowed you to e.g. hack someone by sending them a
specially formed string via IRC if they were running their client in a
shell!)

However it might be the case that everything these days uses ANSI-style
escapes.

On Sat, Feb 28, 2015 at 1:35 PM, Nico Weber [email protected]
wrote:

Right, but is this something one would want to do every time a ninja
command updates? (I honestly don't know.)

If we end up doing #486 #486
(which I currently do want to do, to get colors from clang diagnostics on
Windows), ninja will have some knowledge about ansi codes anyhow, so it
doesn't seem unreasonable for Elide() to learn about them too maybe.

Are there other use cases for zero-width markers other than ansi codes?

—
Reply to this email directly or view it on GitHub
#912 (comment).

[mathstuf]

%0 is output literally with -v, so that path needs to strip things as well.

[mathstuf]

Fixed the -v flag issue.

[mathstuf]

Some parts of this are still necessary. It looks like the width of escape sequences aren't removed when calculating elision. This can cause color leakage if the "clear color" sequence is elided.

[jhasse]

https://github.com/ninja-build/ninja/pull/1454/files#diff-005d118f141929270b3e1c4150887ae2R573 implements the stripping for ElideMiddle. I think we need to adjust this, so that it falls back to result.size() when there aren't any escapes sequences in NINJA_STATUS. This way there wouldn't be a slow down for users who aren't using colors in NINJA_STATUS (besides the one time parsing at the start, which should be negligible).

Also #1210 might be the better solution, to keep Ninja simple.

[mathstuf]

Working on a patch now. It does ANSI detection by "spans" so that there's only a single string.find for escape-less sequences.

[mathstuf]

Also, the linked patch isn't right because it detects elision properly, but doesn't split the string using the escape sequence information to not slice in the middle of one.

[jhasse]

Yeah, getting this right can be tricky, as you also don't want to skip any escape sequences even when they are in the elided area. I've already implemented it in Python here: https://github.com/jhasse/ja/blob/6603076e84083ed9155404268336b5ce5a771bc4/ja/native.py#L244

[mathstuf]

Almost have it. PR up shortly.

[mathstuf]

Closing; this is handled elsewhere (see #1491 and #1494).