OpenType / PostScript font support killed in many apps (FIXED Dec 20, 2012)
(UPDATED repeatedly, first with more details and then because of the Dec 20 fix.)
Was it Fontmageddon? For users who use fonts in some applications (see below), Windows security update KB2753842 of Dec 11, 2012, caused more harm than good. Luckily MS got it fixed and re-released it nine days later. The current version of the patch does NOT have the problem, and can be installed over the original release to fix the problem caused by the original.
Kudos to Microsoft for fixing it quickly and including interested outside parties in testing it. I was able to seed Extensis tech support manager Romeo Fahl with the fixed patch, so we participated in helping verify it worked.
WHAT THE BAD PATCH DID
(1) installing the update breaks some very tiny number of fonts at the system level and for all apps, including potentially malicious fonts. That’s what it was supposed to do. BUT ALSO….
(2) with the original version of the update, for certain apps text set in all PostScript Type 1 (.pfb/.pfm) and OpenType CFF (.oft) fonts became invisible. This can even affect font menus when the app has a WYSIWYG font menu.
FIXING THE PROBLEM
Installing the revised version (2.0) of the patch from Microsoft will fix the problem caused by the original release.
If your computer is part of a domain administered centrally by an IT team, you should alert them that the issue is fixed, so they can decide whether to roll it out now that the patch is safer.
The apps that were especially affected are those that use the GetGlyphOutline() API to grab font outlines of PostScript fonts (both Type 1 pfb/pfm fonts, and OpenType CFF .otf fonts). With the bad version of the patch, that API no longer returned the memory size needed to get the curves, but instead returned a bogus value of zero. This effectively renders some apps unable to render the glyph on screen. At least, at 15 points and higher.
I gather there are other APIs apps can use, but that GetGlyphOutline() works all the way back to XP, unlike the alternatives.
AFFECTED OS VERSIONS AND SOFTWARE
I strongly suspect that in many more applications than those listed, “convert to curves” functions will fail or result in lost text. I also suspect that in most cases where a current version of an application is affected, so are older versions not listed. What we know is that affected OSes and apps included:
- Windows: All desktop and server versions of Windows from XP to Windows 8, it seems.
- PowerPoint, but only in presentation mode (an especially dangerous failure, as a user might think things were fine… until they tried to do an actual presentation)
- QuarkXPress 7, 8, 9.5 (but only affects fonts at 15 pts and larger)
- CorelDRAW X3 to X6. Workaround: view in “draft” mode works because it does not use the problematic API.
- Serif PagePlus
- Adobe Flash (authoring at least, probably not the running of Flash apps?)
- Flexi and SignLab (signmaking apps)
- Avid Marquee (video titling)
- Bentley MicroStation (CAD / information modeling)
- The Secret World (Alternate Reality Game)
- Inkscape (vector drawing)
- Xara Designer Pro X (vector drawing) and possibly other Xara apps
The MS Knowledgebase article has a standard section for “known issues.” On Friday Dec 14, 2012, Microsoft updated it to read: “We are aware of issues related to OpenType Font (OTF) rendering in applications such as PowerPoint on affected versions of Windows that occur after this security update is applied. We are currently investigating these issues and will take appropriate action to address the known issues.”
On Thursday, Dec 20, 2012, Microsoft released version 2.0 of the patch that fixes the problems in the original. The “known issues” section now reads: “The original version of security update 2753842 had an issue related to OpenType Font (OTF) rendering in applications such as PowerPoint on affected versions of Windows. This issue was resolved in the version of this security update that was rereleased on December 20, 2012.”
People keep on sending me links to this article “Boing Boing’s Redesign Uncovers Dark Side of Web Fonts,” about problems Boing Boing had with their new web font implementation. Only thing is, the article has a substantial dose of nonsense mixed in with the perfectly good analysis. I don’t blame the writer, though. This web font stuff is actually really complicated, and information has been hard to come by. Heck, I even briefly forgot a basic point in a first pass at this article. But nonetheless, I am fairly sure that Boing Boing could have fixed their problems easily, if they knew how. Here’s the story:
The background here is that there are new ways of using fonts with web sites, and Boing Boing tried the simplest approach of just hosting a free font on their own web server and pointing at it, but the on-screen results were not as good as expected.
“…the font it settled on — specifically BPreplay — ended up looking terrible for most users.”
“The result was hordes of angry Boing Boing fans complaining that the new headline font was “ugly,” “an abomination” and “plain nasty.” Of course, the culprit wasn’t really the font, but rather how different it looked depending on which browser and operating system the viewer was using.”
FINDING THE CULPRIT?
This is where things get tricky. I will update this post as I learn more. But, as best as I can tell, if that link to the font is right, and the font wasn’t modified by Boing Boing, the culprit really was in large part the font. But first let’s follow the trail of the previous article and make some corrections….
“The problem is that while modern browsers, like the latest versions of Safari, Firefox, Opera and Google Chrome, all support @font-face, the Windows XP operating system often doesn’t have anti-aliasing turned on by default.”
Not true. Anti-aliasing is on by default in XP. What isn’t usually on by default in XP is ClearType, Microsoft’s enhanced anti-aliasing for LCD screens. Sometimes a computer vendor will turn ClearType on for the computers they sell (particularly if they are laptops or come with an LCD monitor).
But what makes this even less meaningful is that (again assuming the linked font is the right one), the font in question is in OpenType CFF format. Such fonts are always anti-aliased in XP, even if you turn off anti-aliasing—the setting change only affects TrueType fonts. Even ClearType doesn’t affect how these fonts are rendered (rasterized) in web browsers, either. Same thing in Vista, and as far as I know in Windows 7 as well.
“The rule, which is still part of CSS3’s draft specification, is also not supported by any version of Internet Explorer. So, as cool as your font might look when properly anti-aliased, on Windows XP it looks, as Rob Beschizza, head of Boing Boing’s redesign puts it, ‘like ass.’”
I’m not sure how those two sentences are related. That’s a mystery to me. XP can get good anti-aliasing as well as Vista, and Internet Explorer is bundled with Vista and Windows 7 as much as with XP. But even taking those points separately….
The @font-face rule was actually in CSS 2 way back when, and removed in CSS 2.1. Internet Explorer has supported @font-face in every version from IE 4 to the current IE 8, but the catch is they support it only with Microsoft’s “.eot” font format (a wrapper around a TrueType font), not with regular desktop fonts. But there’s a good reason for not supporting regular desktop fonts directly: font vendors mostly won’t license their fonts to be stuck “naked” on web servers without any additional protections. Sure, there are free fonts, but, well, we’ll talk about those in a few minutes.
If Boing Boing simply put up the naked .otf font file, and didn’t do anything for Internet Explorer users (and people running older versions of other browsers), then what font actually got displayed to those users would depend on what Boing Boing specified as the fallback fonts and whether the people actually had those fonts (and thanks to Ben Kiel for reminding me of this in passing). Now, if the fallback stack relied on fonts that most XP users would not have, but Vista users would, then there might be a difference that was at least partly based on operating system. But of course it would simply be up to the folks constructing the CSS for the web site to pick reasonable fallback fonts, and not really the fault of the operating system. Perhaps as a first-level fallback from their desired font they specified one of the so-called “ClearType fonts” such as Calibri or Corbel, which are bundled with Windows Vista and Office 2007. Not quite right to blame XP for not having the font, but that would explain how somebody could mislabel it as an XP-specific problem.
Other than that, whether you are using Windows XP or not has little to do with whether or not the font looks “like ass.” Turning on ClearType can further improve rendering for fonts with TrueType outlines, but would have no effect on this particular font. Being on Vista instead of XP would make no difference at all for any web browser rendering OpenType CFF fonts such as this one.
Side note: The font wouldn’t even be seen in Internet Explorer, because Boing Boing didn’t use its .eot format as far as I know. Any further problems in IE would then be due to a poorly-specified font fallback stack in the CSS. However, if they had used .eot and TrueType outlines…. Although XP has ClearType off by default, newer versions of Internet Explorer turn it on just for the browser, so there isn’t much difference between IE rendering of TrueType fonts between XP and Vista.
In researching what went wrong, it doesn’t help that Boing Boing backed off from the font change part of their redesign, so we can’t look at it and test it. Looking at this screenshot, however, shows pretty clearly what was going on. This screen capture was definitely taken on Windows, and the font in question is still anti-aliased. However, it has some pretty crappy artifacts in how it’s rendered on screen, at least some of which are related to it being unhinted. When I see things like a single pixel sticking out of the bottom of a round shaped letter, that’s a dead give-away that the problem is likely hinting (or more accurately, lack thereof).
“Hinting” is essentially extra code in the font that improves its rendering at screen resolution. Apple’s rendering approach largely ignores hinting, but Microsoft’s rendering still uses it a fair bit. Passable hinting can be done automatically by font editing tools, so there’s no real excuse for leaving it out of a font (as with this one). In fact, pretty much every commercial font on the planet is hinted, as are most free fonts. But this isn’t even an average free font. Yes, the terrible spacing is pretty typical for free fonts, but being unhinted is uncommon. Maybe it was converted by somebody else.
Bottom line: the font sucks. This should not be a surprise. Most free fonts do. Don’t get me wrong. There are a few great free fonts out there. But 98% of free fonts suck badly, and maybe about 20% of typically-priced retail fonts suck badly, so set your expectations appropriately.
THE “DIRTY LITTLE SECRET”
But there’s another problem that might have led to complaints and concerns even if the font was made decently, but still in the same format. The “dirty little secret” of the font world is this: Windows GDI rendering of OpenType CFF and “PostScript” Type 1 fonts on screen just sucks, compared to its rendering of TrueType fonts.
Typophiles have long ignored this fact, because in the environments they’ve cared about, Type 1 and OpenType CFF fonts render perfectly well on screen:
- Just about any application on the Mac OS.
- Adobe Acrobat, InDesign and Illustrator, on any platform.
Unfortunately, Windows GDI rendering is what 90% of people see in web browsers and office applications today. (Yes, Safari for Windows also has the option to use its own rendering system, but that’s a tiny minority.) Internet Explorer sidesteps the problem simply by not supporting OpenType CFF fonts in .eot format, only TrueType (though one can convert). But it’s not like .eot ever caught on with web designers, anyway.
The future could improve. There is much better OpenType CFF rendering, even applying ClearType, available for applications using Windows Presentation Foundation and DirectWrite, but very few applications use these modes today, so it is sadly not very relevant… yet. My recollection is that the technical preview of Office 2010 for Windows has dramatically better rendering of OpenType CFF, so perhaps it is coming. Maybe Internet Explorer 9 will get there too, supporting both outline formats in .eot or some new web font format. Perhaps in five years decent support for OpenType CFF rasterization on screen will have reached the strong majority of web browsers….
So for me it’s a toss-up as to whether I blame Windows GDI rendering, or the fact that Boing Boing used a crappy free font (BPreplay) because they couldn’t legally use the retail font they wanted to (VAG Rounded). My first take is that I think Windows GDI just made worse something that would have been a problem anyway. Somebody who knows what they’re doing could spend ten minutes and either fix the font’s hinting in BPreplay or convert it to the TrueType flavor of OpenType—if the license permits it, I’d be happy to try either for them. But then again, maybe the complaint is more about the fallback font, a factor easily controlled by the web site author.
So yes there are some pitfalls. Obviously things would be better if one format worked across all browsers. But there’s also the question of whether one can use the fonts one wants, which tripped up the Boing Boing folks. What happens with that depends on what font vendors decide to do with the fonts they control the licensing for; many foundries are still trying to figure out how to approach the web fonts conundrum. Will they license fonts for use on web servers directly? Will they do so but specify security requirements that can’t be met by sticking regular desktop fonts on web servers, meaning that we’ll have to wait for new web font formats to be widely adopted, such as WOFF? Will they instead rely on a font serving process that involves something centralized, either run by themselves, or by a third party (such as TypeKit or Kernest)?
How exactly this will play out is still TBD today. What I do know for certain is that within a few years, web fonts will be a reality for the average viewer and the average web site. Many or even most web sites will pick specific fonts that aren’t necessarily already on the viewing computer, and those fonts will get used to display the desired text. Font selection will become part of branding for the web the way it has been in print. We’ll also get an explosion of awful font choices on web sites, particularly small personal ones, much like when the masses first got access to a wide variety of fonts they could print on their home computers. But overall, it will be a Good Thing, and I relish the thought of a more typographically rich web world in a year or three.
[Updated for minor clarifications 2009-10-12, reformatted 2009-10-13, added a bit on font fallback 2009-10-13]
A recent thread over on Typophile prompts me to explain why one sometimes sees OpenType CFF fonts that don’t actually support Greek, claiming that they do (by means of the Unicode Range and Codepage bit settings).
Originally, when Adobe converted the Adobe Type Library to OpenType, in the early stages we were thinking we wanted to be as compatible as possible with the Type 1 versions of the fonts.
In the absence of codepage bits and the like in Type 1 fonts, Windows GDI used to do a test for specific codepoints in the font to determine whether given codepages were supported, I believe one codepoint per codepage. I believe the codepoint to determine Greek support was the one for the “mu”… but it was definitely a test for a character that was present in the basic ISO-Adobe character set (now Adobe Latin 1). This even though the character set didn’t really support Greek, it just had a few Greek characters because of their use as math symbols.
So, the Type 1 fonts were (arguably erroneously) detected as supporting Greek by GDI. The idea at the time was that the OpenType fonts with the equivalent character set should behave the exact same way, and therefore the determination of whether to give them the flag saying they support the Greek codepage should be based on the same test… so basically all the fonts would claim to support Greek, even though they really didn’t.
Somehow, even though the idea of near-perfect compatibility between the OpenType fonts and their Type 1 predecessors was abandoned, this principle stuck during the initial conversion of the Adobe Type Library (“Alchemy”). Unicode ranges were set more-or-less in compatibility. Additionally, the AFDKO “makeOTF” tool used to build fonts would automatically do this, unless specifically over-ridden. So you could see third-party fonts doing this as well if they were made with older versions of Adobe tools.
I thought this was a mistake, and eventually convinced other folks of my opinion, so this decision was changed in the revision of the Adobe Type Library (“Facelift”) a couple of years ago, released about October 2007. The AFDKO was changes as well, to match this new preferred behavior.
So, you’ll find that the 1.x version of Adobe Caslon Pro built in 2001 has bits set to claim it supports the Windows Greek codepage and the Greek Unicode range, but this claim was dropped in the 2.0 version of 2007.
Worst Presentation EVAR
I almost didn’t blog about this, because it was probably the most messed-up presentation I’ve done in the last many years. I was trying to do a PDF-based presentation interleaved with a demo in InDesign, but my keyboard stopped working completely when I was in full-screen mode in Acrobat… meaning I also had no way to get out of Acrobat to do the demo! So I had to reboot, re-order my presentation on the fly, and improvise talking through from memory some stuff I had intended to do with accompanying slides, while waiting for my computer to complete the reboot and then for InDesign to launch (which last took 3x as long because I had rebooted while it was running). I also had a cold, so I am clearing my throat every 30 seconds. On top of that, the guy doing the presentation in the next room was REALLY LOUD and somehow his presentation included loud heavy metal music…. Which you can’t hear it on the recording, but I and the audience could hear it very clearly, and it was seriously distracting. Aaargh!
All of which threw me off my pace a bit, even if I seem to be handling it with aplomb on the recording. So even after I’m out of the part where my computer is totally hosed, I’m not at my best.
That being said there’s still some decent stuff in several spots of this AdobeTV recording from Adobe MAX, November 2008. See below for key bits to watch:
- Preserving OpenType alternate glyphs across apps: 5:25 – 8:20 (audio only)
- OpenType typography features: 13:30 – 35:40 (video)
Just don’t expect me to be at my scintillating best on this one, okay?