Hacker News Reader

Comment by mrob | Hacker News Reader

"Switching to the new mode might take some getting used to though, so if you think your fonts are suddenly fat, fuzzy or weird, give your brain some time to adjust."

This is true, and IMO it's the reason font rendering is much less important than people think. When I got a new pair of glasses I disabled font anti-aliasing to make the edges as sharp as possible. I did this so I could more quickly learn to focus my eyes correctly with the new glasses. By the time I had mastered the glasses I was used to the font rendering, and I have not switched it back. I now prefer fonts with no anti-aliasing. If you'd told me in the past that I'd end up preferring that I might not have believed it.

The most legible font is the one you're used to. People can read Blackletter or Spencerian like it's a normal font if they're used to it. I think vector fonts were a mistake. It's a lot of software complexity for very little value.

View on HN

To be fair, the main reason for algorithmic fonts is so that non-latin characters scale up nicely. Storing chinese characters at a high bitmap resolution gets expensive fast, and you'd need several bitmaps for each character to keep them legible at smaller sizes.

Trying to have chinese characters legible at "smaller sizes" is a hopeless cause; they contain much more information than letters do and they consequently need more space. Check out the rendering of the very common character 疆, which HN would like to slot in at basically the same size as the rest of the text. As it shows on my computer, it's legible -- but not correct. (Most egregiously, the horizontal stroke at the bottom of 土 (bottom left quadrant) is missing.) I've seen denser characters show up as blocks of pure black.

Interesting. I do get the bottom stroke of 土 with the Freetype version shipped with Ubunutu 16.04. Most of both 田 is a blur though.

Testing this reminded me that I always read HN two zoom-levels higher than the smallish default.

You might want to give a hidpi display a try (where vector fonts are pretty much essential). All standard ~96 DPI displays look blurry to me after using one.

Though beware of Windows' ClearType, which seems to do a bad job on hidpi: it renders very thin glyphs, and the glyph width discontinuously jumps from 1px to 2px when you change font size. It tries way too hard to snap to pixels. OS X and freetype (tested: 2.6) do a much better job on hidpi.

The problem with Bitmap fonts is that they assume both a pixel density (usually about 96ppi), and a font size. With unicode and smart phones (With up to 6" screens at 4K resolution), it would be impossible to store crisp glyphs for even a fraction of the Basic Multilingual Plane; phone makers still like producing 16GB SKUs! Sure this is one extreme but the benefits of vector fonts can be felt at any ppi with adjustable font sizes.

I'm not saying bitmap fonts don't have their place, but that place is typically old computers or small embedded devices where the complexity of rendering vector fonts is an actual unnecessary performance bottleneck. Vector fonts are easy to use and quick to render, they produce crisp glyphs at virtually (Besides extremely low) resolution and pixel density.

I have a 14" 1080p laptop and so I'm not constantly squinting I use 125% windows scaling and usually about 125% scaling in Google Chrome, although it varies from website to website. With vector fonts this just works, using bitmap fonts would be a lot of ~~software~~ complexity for very little value.

> The most legible font is the one you're used to.

Yes, and in my case on Linux I switched off hinting a long time ago (but retained anti-aliasing)

We even did it on our network at work by default as users liked the fonts when they were more "Mac like".

Personally I have never understood the whole font hinting thing. It seems rather unscientific and the result is that fonts look fundamentally different at slight size differences; seems hinting is part of the problem, not the solution.

> Personally I have never understood the whole font hinting thing. It seems rather unscientific and the result is that fonts look fundamentally different at slight size differences; seems hinting is part of the problem, not the solution.

Hinting exists because displays don't have infinite resolution. Mac-style rendering has always acted like displays have paper-like resolutions, and respects a font's letter shapes more, favoring the original designer's intended font appearance on paper. Windows-style rendering has always acknowledged that displays have pixels, and that smearing fonts across pixels can decrease readability on average-DPI displays. Linux font rendering has historically offered a choice between the two, but typically defaulted to Windows-style rendering.

Yes the most legible font is the one you're used to, but do we want to condition the next generation into thinking low dpi bitmapped fonts are the best? The thing is, we need to set the software defaults to be most pleasing to a population who has never had any prior experience. And IMHO, the print industry has figured that out centuries ago.

but do we want to condition the next generation into thinking low dpi bitmapped fonts are the best?

If it means the software for displaying them can be simpler, maybe yes? Personally I think the X fixed fonts (with a slashed zero) are pretty good... I can read text in that font all day in a terminal and not get tired.

“If I had asked people what they wanted, they would have said faster horses.” - Kurt Cobain.

Wasn't that Henry Ford?

It was probably Michael Scott.

( https://thechive.files.wordpress.com/2015/05/some-things-are... )

Trying to read ClearType for any length of time beyond casual glances actually makes me dizzy and my eyes start watering because of the lack of sharp edges to focus on. Non-subpixel-AA still looks slightly blurry, but I don't get the same "WTF is happening to my eyes" feeling. I prefer non-AA as well, and in the example screenshots posted in the other comment here, it's the font to the left, describing the current settings, that I find most readable.

I wonder if part of the reason for those studies indicating that ClearType "improves reading speed" is because of that - slowing down to focus on the letters feels unpleasant, so people try to scan through the text as quickly as possible.

Maybe you already have but try adjusting your cleartype settings if you have not. It makes a real difference for me.

Just look at any recent Android phone. Font rendering is basically solved, all it took was making displays dense enough. What you are talking about is legacy software that doesn't scale properly.

Little value for the screen when all you want is 80x40, but massively necessary for desktop publishing. Metaphont exists for a reason.

Is that a typo of metafont, or is there another system called metaphont? I did a few web searches and didn't come up with anything.

Is a typo. no idea what my head was doing.

> no idea what my head was doing

Confusing METAFONT with metaphors, presumably.

> I disabled font anti-aliasing to make the edges as sharp as possible. I now prefer fonts with no anti-aliasing

This. A thousand times this.

The sad fact is embedded hand-written quality pixel grid is in old MS core fonts and basically nowhere else to find.

If you have a sufficiently high DPI screen this is definitely the best option.

Font hinting and anti-aliasing is only useful because people have giant ~70-100 dpi screens which look awful without it. On mobile where you have 300-400 DPI it's no issue. Even with 4k desktop monitors with ~200 DPI it seems much less necessary.

I've had the same results as you, though without the glasses. I really don't understand why people prefer such blurry fonts. The biggest practical problem I've had with turning antialiasing off though is badly hinted web fonts. Sometimes bars or stems just outright disappear.

> The most legible font is the one you're used to.

Maybe for you, but don't assume that is true for everybody. It took many months of trying various font rendering configurations (Infinality in ClearType-ish and OSX-ish modes as well as custom modes) before I settled on my current[1] configuration. I'm very familiar with having to spend some time adjusting to different modes, but in my opinion, this the differences in rendering greatly overshadowed the differences from seeing an unusual/new setup.

Perception of text on a screen is affected - at a minimum - by the rendering technique (AA, subpixel-AA, hinting, contrast of the font, contrast of the display, eyesight of the viewer, and the ambient lighting in the room. Maybe in your situation the subtleties of font rendering are not that important. On my monitor, I find [1] to be a lot easier to read than any of the examples (both before and after) that were linked in the announcement.

> very little value

I'm sure most people are fine with sensible defaults, but please, try to remember that "value" is often subjective. People even disagree about which criteria represent high value.

[1] http://i.imgur.com/wMWmNKt.png?1 (esp. the 12px row)

(No, you don't want to see my messy custom fontconfig configuration that overrides a LOT of fo0nts choices and forces them into the handful of fonts that I find easier to read. Yes, that took another few months of experimentation.)

edit:

> vector fonts were a mistake

That depends a lot on what you are doing with them. While I was talking about vector fonts, the amazing font

    -xos4-terminus-medium-r-normal--14-140-72-72-c-80-iso10646-1

is still my favorite. Fortunately, it's easy to use both old-style bitmap fonts and modern fontconfig/freetype rendering in X. Each type of font can be used where they are appropriate. Honorable mention: "Dina".

If you doubt the benefit of a high-quality rendering of a (appropriately hinted) vector font, try loading [2] and [3] (source: a very enlightening article[4]) in separate tabs and switching between them (so the images are in the same position). Image [2] uses modern techniques to render each line 0.1px further right, accumulating 3px over 30 lines. The other image [3] rounds the x offset to the pixel grid. Modern rasterizing techniques can align font rendering to 1/256th of a pixel!

[2] http://www.antigrain.com/research/font_rasterization/sample_...

[3] http://www.antigrain.com/research/font_rasterization/sample_...

[4] http://www.antigrain.com/research/font_rasterization/index.h...

> The most legible font is the one you're used to.

Maybe so, but there's still a need for fonts that are legible before you're used to them, for things like traffic signs.

Reply on news.ycombinator.com