/*
* liberty-tui.c: the ultimate C unlibrary: TUI
*
* Copyright (c) 2016 - 2017, Přemysl Eric Janouch
*
* Permission to use, copy, modify, and/or distribute this software for any
* purpose with or without fee is hereby granted.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY
* SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION
* OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN
* CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*
*/
// This file includes some common stuff to build TUI applications with
#include
// It is surprisingly hard to find a good library to handle Unicode shenanigans,
// and there's enough of those for it to be impractical to reimplement them.
//
// GLib ICU libunistring utf8proc
// Decently sized . . x x
// Grapheme breaks . x . x
// Character width x . x x
// Locale handling . . x .
// Liberal license . x . x
//
// Also note that the ICU API is icky and uses UTF-16 for its primary encoding.
//
// Currently we're chugging along with libunistring but utf8proc seems viable.
// Non-Unicode locales can mostly be handled with simple iconv like in sdtui.
// Similarly grapheme breaks can be guessed at using character width (a basic
// test here is Zalgo text).
//
// None of this is ever going to work too reliably anyway because terminals
// and Unicode don't go awfully well together. In particular, character cell
// devices have some problems with double-wide characters.
#include
#include
#include
#include
// --- Configurable display attributes -----------------------------------------
struct attrs
{
short fg; ///< Foreground colour index
short bg; ///< Background colour index
chtype attrs; ///< Other attributes
};
/// Decode attributes in the value using a subset of the git config format,
/// ignoring all errors since it doesn't affect functionality
static struct attrs
attrs_decode (const char *value)
{
struct strv v = strv_make ();
cstr_split (value, " ", true, &v);
int colors = 0;
struct attrs attrs = { -1, -1, 0 };
for (char **it = v.vector; *it; it++)
{
char *end = NULL;
long n = strtol (*it, &end, 10);
if (*it != end && !*end && n >= SHRT_MIN && n <= SHRT_MAX)
{
if (colors == 0) attrs.fg = n;
if (colors == 1) attrs.bg = n;
colors++;
}
else if (!strcmp (*it, "bold")) attrs.attrs |= A_BOLD;
else if (!strcmp (*it, "dim")) attrs.attrs |= A_DIM;
else if (!strcmp (*it, "ul")) attrs.attrs |= A_UNDERLINE;
else if (!strcmp (*it, "blink")) attrs.attrs |= A_BLINK;
else if (!strcmp (*it, "reverse")) attrs.attrs |= A_REVERSE;
#ifdef A_ITALIC
else if (!strcmp (*it, "italic")) attrs.attrs |= A_ITALIC;
#endif // A_ITALIC
}
strv_free (&v);
return attrs;
}
// --- Terminal output ---------------------------------------------------------
// Necessary abstraction to simplify aligned, formatted character output
// This callback you need to implement in the application
static bool app_is_character_in_locale (ucs4_t ch);
struct row_char
{
ucs4_t c; ///< Unicode codepoint
chtype attrs; ///< Special attributes
int width; ///< How many cells this takes
};
struct row_buffer
{
ARRAY (struct row_char, chars) ///< Characters
int total_width; ///< Total width of all characters
};
static struct row_buffer
row_buffer_make (void)
{
struct row_buffer self = {};
ARRAY_INIT_SIZED (self.chars, 256);
return self;
}
static void
row_buffer_free (struct row_buffer *self)
{
free (self->chars);
}
/// Replace invalid chars and push all codepoints to the array w/ attributes.
static void
row_buffer_append (struct row_buffer *self, const char *str, chtype attrs)
{
// The encoding is only really used internally for some corner cases
const char *encoding = locale_charset ();
// Note that this function is a hotspot, try to keep it decently fast
struct row_char current = { .attrs = attrs };
struct row_char invalid = { .attrs = attrs, .c = '?', .width = 1 };
const uint8_t *next = (const uint8_t *) str;
while ((next = u8_next (¤t.c, next)))
{
current.width = uc_width (current.c, encoding);
if (current.width < 0 || !app_is_character_in_locale (current.c))
current = invalid;
ARRAY_RESERVE (self->chars, 1);
self->chars[self->chars_len++] = current;
self->total_width += current.width;
}
}
static void
row_buffer_append_args (struct row_buffer *self, const char *s, ...)
ATTRIBUTE_SENTINEL;
static void
row_buffer_append_args (struct row_buffer *self, const char *s, ...)
{
va_list ap;
va_start (ap, s);
while (s)
{
row_buffer_append (self, s, va_arg (ap, chtype));
s = va_arg (ap, const char *);
}
va_end (ap);
}
static void
row_buffer_append_buffer (struct row_buffer *self, const struct row_buffer *rb)
{
ARRAY_RESERVE (self->chars, rb->chars_len);
memcpy (self->chars + self->chars_len, rb->chars,
rb->chars_len * sizeof *rb->chars);
self->chars_len += rb->chars_len;
self->total_width += rb->total_width;
}
/// Pop as many codepoints as needed to free up "space" character cells.
/// Given the suffix nature of combining marks, this should work pretty fine.
static int
row_buffer_pop_cells (struct row_buffer *self, int space)
{
int made = 0;
while (self->chars_len && made < space)
made += self->chars[--self->chars_len].width;
self->total_width -= made;
return made;
}
static void
row_buffer_space (struct row_buffer *self, int width, chtype attrs)
{
if (width < 0)
return;
ARRAY_RESERVE (self->chars, (size_t) width);
struct row_char space = { .attrs = attrs, .c = ' ', .width = 1 };
self->total_width += width;
while (width-- > 0)
self->chars[self->chars_len++] = space;
}
static void
row_buffer_ellipsis (struct row_buffer *self, int target)
{
if (self->total_width <= target
|| !row_buffer_pop_cells (self, self->total_width - target))
return;
// We use attributes from the last character we've removed,
// assuming that we don't shrink the array (and there's no real need)
ucs4_t ellipsis = 0x2026; // …
if (app_is_character_in_locale (ellipsis))
{
if (self->total_width >= target)
row_buffer_pop_cells (self, 1);
if (self->total_width + 1 <= target)
row_buffer_append (self, "…", self->chars[self->chars_len].attrs);
}
else if (target >= 3)
{
if (self->total_width >= target)
row_buffer_pop_cells (self, 3);
if (self->total_width + 3 <= target)
row_buffer_append (self, "...", self->chars[self->chars_len].attrs);
}
}
static void
row_buffer_align (struct row_buffer *self, int target, chtype attrs)
{
row_buffer_ellipsis (self, target);
row_buffer_space (self, target - self->total_width, attrs);
}
static void
row_buffer_print (uint32_t *ucs4, chtype attrs)
{
// This assumes that we can reset the attribute set without consequences
char *str = u32_strconv_to_locale (ucs4);
if (str)
{
attrset (attrs);
addstr (str);
attrset (0);
free (str);
}
}
static void
row_buffer_flush (struct row_buffer *self)
{
if (!self->chars_len)
return;
// We only NUL-terminate the chunks because of the libunistring API
uint32_t chunk[self->chars_len + 1], *insertion_point = chunk;
for (size_t i = 0; i < self->chars_len; i++)
{
struct row_char *iter = self->chars + i;
if (i && iter[0].attrs != iter[-1].attrs)
{
row_buffer_print (chunk, iter[-1].attrs);
insertion_point = chunk;
}
*insertion_point++ = iter->c;
*insertion_point = 0;
}
row_buffer_print (chunk, self->chars[self->chars_len - 1].attrs);
}