On Tue, Feb 15, 2022, 6:57 PM Philippe Mathieu-Daudé <[email protected]> wrote:
> On 16/2/22 00:53, John Snow wrote: > > On Tue, Feb 15, 2022 at 5:55 PM Eric Blake <[email protected]> wrote: > >> > >> On Tue, Feb 15, 2022 at 05:08:50PM -0500, John Snow wrote: > >>>>>> print(enboxify(msg, width=72, name="commit message")) > >>> ┏━ commit message > ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┓ > >>> ┃ enboxify() takes a chunk of text and wraps it in a text art box that > ┃ > >>> ┃ adheres to a specified width. An optional title label may be given, > ┃ > >>> ┃ and any of the individual glyphs used to draw the box may be > ┃ > >> > >> Why do these two lines have a leading space, > >> > >>> ┃ replaced or specified as well. > ┃ > >> > >> but this one doesn't? It must be an off-by-one corner case when your > >> choice of space to wrap on is exactly at the wrap column. > >> > > > > Right, you're probably witnessing the right-pad *and* the actual space. > > > >>> > ┗━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┛ > >>> > >>> Signed-off-by: John Snow <[email protected]> > >>> --- > >>> python/qemu/utils/__init__.py | 58 > +++++++++++++++++++++++++++++++++++ > >>> 1 file changed, 58 insertions(+) > > >>> + def _wrap(line: str) -> str: > >>> + return os.linesep.join([ > >>> + wrapped_line.ljust(lwidth) + suffix > >>> + for wrapped_line in textwrap.wrap( > >>> + line, width=lwidth, initial_indent=prefix, > >>> + subsequent_indent=prefix, > replace_whitespace=False, > >>> + drop_whitespace=False, break_on_hyphens=False) > >> > >> Always nice when someone else has written the cool library function to > >> do all the hard work for you ;) But this is probably where you have > the off-by-one I called out above. > >> > > > > Yeah, I just didn't want it to eat multiple spaces if they were > > present -- I wanted it to reproduce them faithfully. The tradeoff is > > some silliness near the margins. > > > > Realistically, if I want something any better than what I've done > > here, I should find a library to do it for me instead -- but for the > > sake of highlighting some important information, this may be > > just-enough-juice. > > 's/^┃ /┃ /' on top ;D > I have to admit that this function is actually very fragile. Last night, I did some reading on unicode and emoji encodings and discovered that it's *basically impossible* to predict the "visual width" of a sequence of unicode codepoints. So, this function as written will only really work if we stick to single-codepoint glyphs that can be rendered 1:1 in a monospace font. I could probably improve it to work with "some" (but certainly not all) wide glyphs and emoji, but it's a very complex topic and far outside my specialty. Support for multi-codepoint narrow/halfwidth glyphs is also an issue. (This affects some Latin characters outside of ascii if they are encoded using combining codepoints.) (See https://hsivonen.fi/string-length/ ... It's nasty.) So I must admit that this function has some very serious limitations to it. I want to explain why I wrote it, though. First: Tracebacks make people's eyes cross over. It's a very long sequence of mumbo jumbo that most people don't read, because it's program debug information. I don't blame them. Setting apart the error summary visually is a helpful tool for drawing one's eyes to the most critical pieces of information. Second: In my AQMP library, I use the ascii vertical bar | as a left-hand border decoration to provide a kind of visual quoting mechanism to illustrate in the logfile which subsequent confusing lines of jargon belong to the same log entry. I really like this formatting mechanism, but... Third: If a line of text becomes so long that it wraps in your terminal, the visual quote mechanism breaks, making the output messy and hard to read. Forcibly re-wrapping the text in a virtual box is a necessary mechanism to preserve readability in this circumstance - the lines from qemu-img et al may be much wider than your terminal column width. And so, I drew a box instead of just a left border, because I needed to re-wrap the text anyway. Visually, I believed it to help explain that the output was being re-formatted to fit in a certain dimensionality. Unfortunately, it's inadequate. So ... what to do. (1) I can just remove the right margin decoration and call the function visual_quote or something. If any of the lines get too "long" because of emoji/日本語, it MAY break the indent line, but occasional uses of one or two wide characters probably won't cause wrapping that breaks the "visual quote line" on a terminal with at least 85 columns. Essentially it'd still be broken, but without a solid right border it'd be harder to notice *small* breakages. (2) If there is a genuine interest in using visual highlighting techniques to make iotest failures easier to diagnose (and making sure it is properly multilingual), I could use the urwid helper library to estimate visual text width to make drawing terminal boxes more resilient than what I could do on my own power. Downside is a new third party dependency. I already use urwid for the aqmp tui that we're working on, but it's remained an optional dependency so far. (3) I can take a swing at improving this text decoration utility and having it account for the most basic cases. East Asian language support is a low hanging fruit, though I have only rudimentary familiarity with Hangul. (And virtually no exposure to Thai or other south-eastern Asian scripts.) (4) Just leave it alone for now, don't you have IDE/FDC patches to work on? Sigh. The punishment for trying to do something nice is swift. --js >
