Web lists-archives.com

Re: make compilation not so gray

[Warning: a long-winded rant about terminals.  You should have known better
than to get me onto this particular topic.]

On Sat, May 26, 2018 at 12:57:39PM +1200, Ben Caradoc-Davies wrote:
> On 26/05/18 12:28, Ben Caradoc-Davies wrote:
> > On 26/05/18 00:21, Adam Borowski wrote:
> > > It might be good to pipe output through ansi2txt
> > Thank you so much for alerting me to ansi2txt. Given the miserable
> > failure of some command line tool authors to contemplate terminal
> > background colours other than black, I have had to endure all manner of
> > workarounds. One investigation revealed ANSI sequences hardcoded in the
> > source with no way to turn them off. The horror.
> > In my view, colour output should not be the default. Anyone who causes
> > yellow or light green or, worst of all, white text to be displayed on a
> > white background should be dealt with by the a11y police.
> Exhibit A: dpkg-buildpackage (attached).

As Stuart mentioned, this is entirely the fault of your terminal, not of
dpkg.  Is this a preset, or something you personally configured?  If the
former, please file a bug report -- if the latter, please fix your setup. 
With only 16 basic colors out of 24-bit color space, there's plenty of
palettes that make every combination very easy to read.  Your screenshot
uses #ffff55 for yellow -- human vision makes the blue component really
insignificant, thus you'd need to put it down further.

I don't know what a proper nice value would be -- we'd need to ask someone
with knowledge about color theory, artistic skills and what not; in
"ansi2html -w" (ie, in "on-white" mode I consider a burnination-worthy
heresy) I fudged yellow all the way down to #cccc00 and brown to #806000.

The standard defines that \e[33;1m means "yellow" without specifying the
exact shade; most terminals offer a bunch of presets you can choose from.
If a preset selected by default makes the text unreadable, it's a fault
of that preset.

> Another offender is megatools, because who doesn't like progress bars with
> hardcoded ANSI sequences for white and yellow foreground regardless of
> background?:
> https://github.com/megous/megatools/blob/master/lib/tools.c#L270
> https://github.com/megous/megatools/blob/master/lib/tools.h#L60

Yellow is perfectly fine, per the reasoning above.

I agree that white should not be used.  Unless the program sets both
foreground and background, it must not assume on-black vs on-white.  Thus,
the appropriate attributes that auto-adjust are:
* normal (39, 22 (usually set via 0)): default foreground color
* bright/bold (1)
* reverse (7)

I don't use that Mega thingy so I'd prefer not to send untested patches, but
I'd suggest you to change and submit this.  Hard-coding the ANSI codes,
though, is the way to go -- vt100 won.  Since vt100 took the world of
terminals with a storm, it defined the SGR syntax and that unknown SGR codes
can be cleanly ignored.  Thus, even if a terminal doesn't support color
(like vt100 itself!), using such a code unconditionally is safe.

> What's a termcap?

A technique and database that has been left unmaintained since mid-1980s.
It never worked right.  Not during the great Unix Wars, not today.  The
design is outright wrong: how exactly the definition for your swanky new
terminal would be known by a remote machine you log to in if not only it
doesn't have swanky-new-terminal package locally installed but it did not
even exist when it was released (it's a stable or oldstable server)?
Some popular transports for a text console, such as a serial link, don't
provide a way to send metadata -- but those which do, like ssh, don't
send any termcap definitions either.

Take a look at what different terminals say.  Every terminal other than
linux (vt) and rxvt identify themselves as just "xterm", despite xterm vs
vte9 vs vte-2.91 vs konsole vs qterminal vs putty vs ... all having pretty
different behaviour.  Ok, recently all of these started changing to
xterm-256color because ncurses' upstream still believes in termcap, despite
its syntax being incapable of handling 256 or 24-bit color mode sanely.

If you want a recent example, WSL's console has just wontfixed a request to
stop pretending to be "xterm": https://github.com/Microsoft/WSL/issues/52
after two years of trying.

> An appropriate penalty for careless use of ANSI sequences would be for
> offenders to be required to include this in their ~/.bashrc for two years:
> echo -en "\e]10;black\x7\e]11;black\x7"

Er... you want to print a _beep_?  What?

This is a common antipattern: one specification for OSC codes was
egregiously wrong, with gems such as allowing \n in the middle (acting
normally without interrupting the command), and having most byte values
outside of printable 7-bit ASCII terminate the command.  This includes
values with a high bit set (window titles in non-ASCII anyone?).  And
somehow, some then-popular terminals parsed invalid values as "terminate and
accept" instead of "terminate and abort", thus beep got cargo-culted as
a terminator for OSC commands in some docs, despite being one of worst
possible characters for this role.

You really want "\e\\" (ESC '\') instead.

Then, using some X database color names here is another shootable offense. 
How do you expect an embedded terminal, a terminal that's a part of
something like the kernel, or pretty much anything but real xterm linked to
historic X libs to carry a database to know what "black" or "BlanchedAlmond"
should map to?

So yeah... your penalty for careless use of ANSI sequences needs more care.

⢀⣴⠾⠻⢶⣦⠀ .globl _start↵.data↵rc: .ascii "/etc/init.d/rcS\0"↵.text↵_start
⣾⠁⢰⠒⠀⣿⡁ mov $57,%rax↵syscall↵cmp $0,%rax↵jne child↵parent:↵mov $61,%rax
⢿⡄⠘⠷⠚⠋⠀ mov $-1,%rdi↵xor %rsi,%rsi↵xor %rdx,%rdx↵syscall↵jmp parent↵child:
⠈⠳⣄⠀⠀⠀⠀ mov $59,%rax↵mov $rc,%rdi↵xor %rsi,%rsi↵xor %rdx,%rdx↵syscall