Tk Source Code

It seems that the font names Tk provides for compatability are blocking actual, installed system fonts. From the manpage for the -family option to the [font] command:

The case-insensitive font family name. Tk guarantees to support the font families named Courier (a monospaced “typewriter” font), Times (a serifed “newspaper” font), and Helvetica (a sans-serif “European” font). The most closely matching native font family will automatically be substituted when one of the above font families is used. The name may also be the name of a native, platform-specific font family; in that case it will work as desired on one platform but may not display correctly on other platforms. If the family is unspecified or unrecognized, a platform-specific default font will be chosen.

I have a font installed actually called "Courier", but it seems to be impossible to use it in Tk:

(Mike) 17 % font actual Courier
-family {Courier New} -size 12 -weight normal -slant roman -underline 0 -overstrike 0
(Mike) 18 % font actual [list -family Courier]
-family {Courier New} -size 12 -weight normal -slant roman -underline 0 -overstrike 0

Tk's compatability name seems to be checked before actual installed fonts, so any attempt to access it's being mapped to Courier New. Especially annoying since the [tk fontchooser], which uses the built in Windows font dialog, does show Courier and Courier New as two distinct and separate fonts.

Since Tk provides a number of configurable named fonts these days (TkDefaultFont, etc), perhaps these three should be moved to there (right now they seem to be special, included neither in [font names] nor [font families]), so they can at least be manually edited or removed by applications if desired?

A second warning sent to Tcl Core list did not trigger any reaction, and no TIP was requested, therefore this is now merged to trunk (only): [678fdae0]

And on Linux (Debian 8), the changes are fine too (tests results in trunk and in the bugfix branch show the same results, modulo unstable tests).

I have now also checked that the proposed changes in the bugfix branch are fine for OS X too.

I have committed the proposed change in branch bug-c0dbdd3ff3, and have fixed all failing tests (checked this on WIn only so far).

Ahh, I see. Sounds like it's probably a matter of going through all the test cases and changing the fonts being used in the tests from 'times' to 'Times New Roman', etc, so they still match up with previous behaviour?

I have attached the Tk test suite results on windows Vista, in Trunk, after removal of the lines ruling out some fonts in FalilyExists().

A few failures are due to what you mention, but most are due to checks based on measuring some distance on screen, which is very sensitive on the font. OK, yes, probably poorly written tests. But this is a lot of work to correct.

Wow. For such a tiny and unremarkable change, that is an impressive number of failures. ;) What are the failures, out of curiousity? If you don't have matching fonts installed, all three should still exist the same way using the fallbacks, shouldn't they? I'm wondering if some of the failures are because you do have a Courier font, and it's checking for [font actual "Courier" -family] being "Courier New", etc. (Though even if that is the case, it surely wouldn't account for that many failures...)

I received no answer when I prompted the Tcl Core list for opinions on removing this.

So I gave it a try.

Hmpf. In trunk, on Windows, removing these lines makes 135 tests of the Tk test suite newly fail !!!

It looks like the reason it's there is that whoever did it thought the fonts were ugly, from the comments? That seems a really bad idea to block access to those specific fonts at that level, especially since they're not blocked from other places in Tk (such as the built-in Windows font dialog, so people can still try and select it should they wish to and then end up with what seems to be a bug).

Back in 1999, Courier at least was likely the crappy version that pixelated, not a proper truetype font (though I've seen plenty of perfectly good fonts called Times and Helvetica on different Windows machines over the years).

This just seems really arbitrary, and something much better left to individual program writers to me.

Sorry, my link below was in fact not relevant. A less superficial analysis of what is happening when calling font actual Courier (on Windows, which is the case for the present report) is as follows:

Tk_FontObjCmd(): case FONT_ACTUAL looks for the font named "Courier" by calling Tk_AllocFontFromObj().

This call first looks (TkpGetNativeFont()) for a native font of this name. There is none, therefore it tries to find the closest matching font by calling TkpGetFontFromAttributes().

TkpGetFontFromAttributes() first tries the font name, "Courier", and looks if a font family or an alias exists: FamilyOrAliasExists(), which first calls FamilyExists().

And in FamilyExists() there is an intentional rule out of "Courier" font.

Removing those lines in the source code makes font actual Courier return:

 -family Courier -size 12 -weight normal -slant roman -underline 0 -overstrike 0

This has been in Tk since 1999, and the reason why it's there looks valid to me.

Huh, looks like there are a few others there (mincho, gothic, and dingbats) that work in the same way but are also undocumented? Candidates for the same treatment. (Is it worth raising a separate ticket about them being undocumented?)

Perhaps this is relevant.