I made a patch that improves the performance of font caching mechanism.
This is based on a funny behaviour of FontConfig: it was handling
FcCharSet in a somewhat unexpected way.
So, we are currently adding "a character" to a new FcCharSet, and then
add it to a FcPattern. However, if we toss the FcPattern to FontConfig,
it loads the entire language(charset) that contains the character we
gave. That is, we don't always have to load a new font for each unknown
character. Instead, we can reused cached fonts, and this significantly
reduces the number of calls to extremely slow FontConfig matching
functions.
One more thing. I found that, in libXft, there's a function called
XftCharExists. XftCharIndex internally calls this function, and
does more stuffs if the character does exist. Since the returned index
is never used in st, we should call XftCharExists instead of
XftCharIndex. Please note that I already made this change in the patch.
To be more specific, now tty creation is delayed until X window is
actually mapped; last ConfigureNotify before mapping determines
initial tty size.
Please report problems if there are any.
There were two problems with match denfinition.
1) There was a forward declaration in the form:
static inline bool match(uint, uint);
but later the function was defined as:
inline bool
match(uint mask, uint state) {
This causes that there were two different functions in the code, one local
and inline, and other inline but extern. All was working without problems
due to we were using -Os, and the compiler was using the extern definition
and it was no expanding the static declaration. If you removed the -Os flag,
then you got linker errors due it was no able to find the static definition
of the static declaration.
2) The mask checking was incorrect because we were doing the test:
(state & mask) != state
and this test only was saying that at least all the enabled bits of state
were enabled also in mask, but no all the possible bits in mask. This was
the origin of the bug reported by Xavier Cartron, where he said it was
possible activated some shortcuts with some of the modifiers defined in the
config.h file.
draw is the function which update the Xwindow with the information st has,
and it is designed in a way that it must be called once in the main loop
(run function), and calling it in other places it is a waste of time.
The way st knows if there is a selection activated is checking if sel.ob.x
is equal to -1. In some parts of the code the way of disabling the selection
was only setting it to -1, but after it you can't be sure if the selection
is clearing from the terminal representation, because it is necessary mark
all the lines affected by the selection as dirty. Already there is a functon
which perform this task, selclear.
We're now clearing empty areas with spaces, so there is no point to check
if character contains non-empty string.
Signed-off-by: Christoph Lohmann <20h@r-36.net>
The commit b78c5085f7 changed the st behaviour enabling BCE capability,
that means erase regions using background color. Problem comes when you
clear a region with a selection, because in this case the real mode of the
Glyph is not the value of term.line[y][x], due in drawregion we had enabled
the ATTR_REVERSE bit.
Signed-off-by: Christoph Lohmann <20h@r-36.net>
Now double-click+dragging automatically snaps both ends to word boundaries
(unless on series of spaces), and triple-click selects whole lines.
As a side effect, snapping now occurs on button press, not button release
like it previously was, but I hope that won't be inconvenient for anyone.
Signed-off-by: Christoph Lohmann <20h@r-36.net>
Now, when you are selecting a region, you will get all empty lines that happen
to be in it, including trailing ones. Last line terminator is omitted as it previously
was, though.
Signed-off-by: Christoph Lohmann <20h@r-36.net>
