Tk Source Code

Named fonts are not configured properly when the font name is used in a widget prior to the creation of the named font by the font create command as shown in the following snippet.

proc step1 {} {
    after 1000 step2
    ttk::style configure TLabel -font ANamedFont
    label .l1 -text "A normal label" -font ANamedFont
    ttk::label .l2 -text "A ttk::label"
    pack .l1 .l2
}
proc step2 {} {
    after 1000 step3
    font create ANamedFont -family Helvetica -size 30
}
proc step3 {} {
    ttk::label .l3 -text "Another ttk::label"
    pack .l3
}
step1

The expected outcome would be the first two labels having Helvetica 30 after a second, and the third label having that same font after two seconds. However, no font change happens both on Win32 and POSIX (Mac untested), i.e. the system chosen fallback font remains in use. In contrast, this snippet does what one expects:

proc step1 {} {
    after 1000 step2
    ttk::style configure TLabel -font ANamedFont
    font create ANamedFont -family Helvetica -size 30
    label .l1 -text "A normal label" -font ANamedFont
    ttk::label .l2 -text "A ttk::label"
    pack .l1 .l2
}
proc step2 {} {
    after 1000 step3
    font delete ANamedFont
    font create ANamedFont -family Courier -size 30
}
proc step3 {} {
    ttk::label .l3 -text "Another ttk::label"
    pack .l3
}
step1

So there's some confusion in the font subsystem regarding caching of named and unnamed fonts and when to perform updates. Therefore I attach a proposed fix as a patch against core-8-6-branch.

Harald,

while I can live with the test, doing this in another interp serves a purpose: to get a ultra clean (Mr Proper) start condition for the test, since that interp loads and initializes its own little Tk world. Just a thought.

Without further instrumentation it might be possible to test it somehow unsharp with this approach

test font-x.x {Named font late creation} -setup {
    catch {interp delete foo}
} -body {
    interp create foo
    foo eval {
        load {} Tk
        font create F1 -family Helvetica -size 100
        label .l -font F1 -text LARGE
        pack .l
        update
        set W1 [winfo width .l]
        destroy .l
        font delete F1
        label .l -font F2 -text LARGE
        pack .l
        update
        font create F2 -family Helvetica -size 100
        update
        set W2 [winfo width .l]
        destroy .l
        font delete F2
    }
    set result [foo eval {expr {$W1 == $W2}}]
    foo eval {destroy .}
    interp delete foo
} -result 1

but this is a sketch only. Please refine.

Harald,

thanks for testing. Indeed is this problem as old as Tk 8.0 as far as can be seen in tkFont.c. In the second snippet of the ticket, the font delete is essential, since it proves that on re-creation of the same named font an update is performed. But the fix addresses the initial case, i.e. named font non-existent, font subsystem uses a fallback, and after that the named font is created in the expectation to be immediately used.