Hi,
the following patch had been posted around spring 2005, but has not since
been included. It is about that hitting the keys <Compose><o><e> for example
did not produce the utf8 sequence for o-with-diaeresis, but the ASCII/ISO8859-1
char for o-w-diaeresis, as I seem to recall.
http://groups.google.de/group/linux.kernel/browse_thread/thread/e60f286969e83d99/55688a5aab326aa7?tvc=2&q=chris+(+%22utf-8%22+OR+%22utf8%22+)+console#55688a5aab326aa7
I am posting an updated version that I hope applies to 2.6.15-rc6.
Not sure what the correct X-Y-by: procedure is for this, as it originally
is not my work (see URL).
diff -Ppru linux-2.6.15-rc6-20051219230006/drivers/char/consolemap.c linux-2.6-AS22/drivers/char/consolemap.c
--- linux-2.6.15-rc6-20051219230006/drivers/char/consolemap.c 2005-12-11 13:42:23.000000000 +0100
+++ linux-2.6-AS22/drivers/char/consolemap.c 2005-12-19 21:53:25.000000000 +0100
@@ -178,6 +178,7 @@ struct uni_pagedir {
unsigned long refcount;
unsigned long sum;
unsigned char *inverse_translations[4];
+ u16 *inverse_trans_unicode;
int readonly;
};
@@ -208,6 +209,40 @@ static void set_inverse_transl(struct vc
}
}
+static void set_inverse_trans_unicode(struct vc_data *conp,
+ struct uni_pagedir *p)
+{
+ int i, j, k, glyph;
+ u16 **p1, *p2;
+ u16 *q;
+
+ q = p->inverse_trans_unicode;
+ if (!q) {
+ q = p->inverse_trans_unicode =
+ kmalloc(MAX_GLYPH * sizeof(u16), GFP_KERNEL);
+ if (!q)
+ return;
+ }
+ memset(q, 0, MAX_GLYPH * sizeof(u16));
+
+ for (i = 0; i < 32; i++) {
+ p1 = p->uni_pgdir[i];
+ if (!p1)
+ continue;
+ for (j = 0; j < 32; j++) {
+ p2 = p1[j];
+ if (!p2)
+ continue;
+ for (k = 0; k < 64; k++) {
+ glyph = p2[k];
+ if (glyph >= 0 && glyph < MAX_GLYPH
+ && q[glyph] < 32)
+ q[glyph] = (i << 11) + (j << 6) + k;
+ }
+ }
+ }
+}
+
unsigned short *set_translate(int m, struct vc_data *vc)
{
inv_translate[vc->vc_num] = m;
@@ -218,19 +253,29 @@ unsigned short *set_translate(int m, str
* Inverse translation is impossible for several reasons:
* 1. The font<->character maps are not 1-1.
* 2. The text may have been written while a different translation map
- * was active, or using Unicode.
+ * was active.
* Still, it is now possible to a certain extent to cut and paste non-ASCII.
*/
-unsigned char inverse_translate(struct vc_data *conp, int glyph)
+u16 inverse_translate(struct vc_data *conp, int glyph, int use_unicode)
{
struct uni_pagedir *p;
+ int m;
if (glyph < 0 || glyph >= MAX_GLYPH)
return 0;
- else if (!(p = (struct uni_pagedir *)*conp->vc_uni_pagedir_loc) ||
- !p->inverse_translations[inv_translate[conp->vc_num]])
+ else if (!(p = (struct uni_pagedir *)*conp->vc_uni_pagedir_loc))
return glyph;
- else
- return p->inverse_translations[inv_translate[conp->vc_num]][glyph];
+ else if (use_unicode) {
+ if (!p->inverse_trans_unicode)
+ return glyph;
+ else
+ return p->inverse_trans_unicode[glyph];
+ } else {
+ m = inv_translate[conp->vc_num];
+ if (!p->inverse_translations[m])
+ return glyph;
+ else
+ return p->inverse_translations[m][glyph];
+ }
}
static void update_user_maps(void)
@@ -354,6 +399,10 @@ static void con_release_unimap(struct un
kfree(p->inverse_translations[i]);
p->inverse_translations[i] = NULL;
}
+ if (p->inverse_trans_unicode) {
+ kfree(p->inverse_trans_unicode);
+ p->inverse_trans_unicode = NULL;
+ }
}
void con_free_unimap(struct vc_data *vc)
@@ -512,6 +561,7 @@ int con_set_unimap(struct vc_data *vc, u
for (i = 0; i <= 3; i++)
set_inverse_transl(vc, p, i); /* Update all inverse translations */
+ set_inverse_trans_unicode(vc, p);
return err;
}
@@ -562,6 +612,7 @@ int con_set_default_unimap(struct vc_dat
for (i = 0; i <= 3; i++)
set_inverse_transl(vc, p, i); /* Update all inverse translations */
+ set_inverse_trans_unicode(vc, p);
dflt = p;
return err;
}
@@ -618,6 +669,19 @@ void con_protect_unimap(struct vc_data *
p->readonly = rdonly;
}
+/* may be called during an interrupt */
+u32 conv_8bit_to_uni(unsigned char c)
+{
+ /*
+ * Always use USER_MAP. This function is used by the keyboard,
+ * which shouldn't be affected by G0/G1 switching, etc.
+ * If the user map still contains default values, i.e. the
+ * direct-to-font mapping, then assume user is using Latin1.
+ */
+ unsigned short uni = translations[USER_MAP][c];
+ return uni == (0xf000 | c) ? c : uni;
+}
+
int
conv_uni_to_pc(struct vc_data *conp, long ucs)
{
diff -Ppru linux-2.6.15-rc6-20051219230006/drivers/char/keyboard.c linux-2.6-AS22/drivers/char/keyboard.c
--- linux-2.6.15-rc6-20051219230006/drivers/char/keyboard.c 2005-12-19 21:53:05.000000000 +0100
+++ linux-2.6-AS22/drivers/char/keyboard.c 2005-12-19 21:53:25.000000000 +0100
@@ -34,6 +34,7 @@
#include <linux/init.h>
#include <linux/slab.h>
+#include <linux/consolemap.h>
#include <linux/kbd_kern.h>
#include <linux/kbd_diacr.h>
#include <linux/vt_kern.h>
@@ -353,6 +354,15 @@ static void to_utf8(struct vc_data *vc,
}
}
+static void put_8bit(struct vc_data *vc, u8 c)
+{
+ if (kbd->kbdmode != VC_UNICODE || c < 32 || c == 127)
+ /* Don't translate control chars */
+ put_queue(vc, c);
+ else
+ to_utf8(vc, conv_8bit_to_uni(c));
+}
+
/*
* Called after returning from RAW mode or when changing consoles - recompute
* shift_down[] and shift_state from key_down[] maybe called when keymap is
@@ -413,7 +423,7 @@ static unsigned char handle_diacr(struct
if (ch == ' ' || ch == d)
return d;
- put_queue(vc, d);
+ put_8bit(vc, d);
return ch;
}
@@ -423,7 +433,7 @@ static unsigned char handle_diacr(struct
static void fn_enter(struct vc_data *vc, struct pt_regs *regs)
{
if (diacr) {
- put_queue(vc, diacr);
+ put_8bit(vc, diacr);
diacr = 0;
}
put_queue(vc, 13);
@@ -632,7 +642,7 @@ static void k_self(struct vc_data *vc, u
diacr = value;
return;
}
- put_queue(vc, value);
+ put_8bit(vc, value);
}
/*
diff -Ppru linux-2.6.15-rc6-20051219230006/drivers/char/selection.c linux-2.6-AS22/drivers/char/selection.c
--- linux-2.6.15-rc6-20051219230006/drivers/char/selection.c 2005-12-11 13:42:24.000000000 +0100
+++ linux-2.6-AS22/drivers/char/selection.c 2005-12-19 21:53:25.000000000 +0100
@@ -20,6 +20,7 @@
#include <asm/uaccess.h>
+#include <linux/kbd_kern.h>
#include <linux/vt_kern.h>
#include <linux/consolemap.h>
#include <linux/selection.h>
@@ -34,6 +35,7 @@ extern void poke_blanked_console(void);
/* Variables for selection control. */
/* Use a dynamic buffer, instead of static (Dec 1994) */
struct vc_data *sel_cons; /* must not be disallocated */
+static int use_unicode;
static volatile int sel_start = -1; /* cleared by clear_selection */
static int sel_end;
static int sel_buffer_lth;
@@ -54,10 +56,8 @@ static inline void highlight_pointer(con
complement_pos(sel_cons, where);
}
-static unsigned char
-sel_pos(int n)
-{
- return inverse_translate(sel_cons, screen_glyph(sel_cons, n));
+static u16 sel_pos(int n) {
+ return inverse_translate(sel_cons, screen_glyph(sel_cons, n), use_unicode);
}
/* remove the current selection highlight, if any,
@@ -86,8 +86,8 @@ static u32 inwordLut[8]={
0xFF7FFFFF /* latin-1 accented letters, not division sign */
};
-static inline int inword(const unsigned char c) {
- return ( inwordLut[c>>5] >> (c & 0x1F) ) & 1;
+static inline int inword(const u16 c) {
+ return c > 0xff || ((inwordLut[c >> 5] >> (c & 0x1F)) & 1);
}
/* set inwordLut contents. Invoked by ioctl(). */
@@ -108,13 +108,35 @@ static inline unsigned short limit(const
return (v > u) ? u : v;
}
+/* stores the char in UTF8 and returns the number of bytes used (1-3) */
+int store_utf8(u16 c, char *p) {
+ if (c < 0x80) {
+ /* 0******* */
+ p[0] = c;
+ return 1;
+ } else if (c < 0x800) {
+ /* 110***** 10****** */
+ p[0] = 0xc0 | (c >> 6);
+ p[1] = 0x80 | (c & 0x3f);
+ return 2;
+ } else {
+ /* 1110**** 10****** 10****** */
+ p[0] = 0xe0 | (c >> 12);
+ p[1] = 0x80 | ((c >> 6) & 0x3f);
+ p[2] = 0x80 | (c & 0x3f);
+ return 3;
+ }
+}
+
/* set the current selection. Invoked by ioctl() or by kernel code. */
int set_selection(const struct tiocl_selection __user *sel, struct tty_struct *tty)
{
struct vc_data *vc = vc_cons[fg_console].d;
int sel_mode, new_sel_start, new_sel_end, spc;
char *bp, *obp;
- int i, ps, pe;
+ int i, ps, pe, multiplier;
+ u16 c;
+ struct kbd_struct *kbd = kbd_table + fg_console;
poke_blanked_console();
@@ -158,6 +180,7 @@ int set_selection(const struct tiocl_sel
clear_selection();
sel_cons = vc_cons[fg_console].d;
}
+ use_unicode = kbd && kbd->kbdmode == VC_UNICODE;
switch (sel_mode)
{
@@ -240,7 +263,8 @@ int set_selection(const struct tiocl_sel
sel_end = new_sel_end;
/* Allocate a new buffer before freeing the old one ... */
- bp = kmalloc((sel_end-sel_start)/2+1, GFP_KERNEL);
+ multiplier = use_unicode ? 3 : 1; /* chars can take up to 3 bytes */
+ bp = kmalloc((sel_end-sel_start)/2*multiplier+1, GFP_KERNEL);
if (!bp) {
printk(KERN_WARNING "selection: kmalloc() failed\n");
clear_selection();
@@ -251,8 +275,12 @@ int set_selection(const struct tiocl_sel
obp = bp;
for (i = sel_start; i <= sel_end; i += 2) {
- *bp = sel_pos(i);
- if (!isspace(*bp++))
+ c = sel_pos(i);
+ if (use_unicode)
+ bp += store_utf8(c, bp);
+ else
+ *bp++ = c;
+ if (!isspace(c))
obp = bp;
if (! ((i + 2) % vc->vc_size_row)) {
/* strip trailing blanks from line and add newline,
diff -Ppru linux-2.6.15-rc6-20051219230006/include/linux/consolemap.h linux-2.6-AS22/include/linux/consolemap.h
--- linux-2.6.15-rc6-20051219230006/include/linux/consolemap.h 2005-10-28 02:02:08.000000000 +0200
+++ linux-2.6-AS22/include/linux/consolemap.h 2005-12-19 21:53:25.000000000 +0100
@@ -10,6 +10,7 @@
struct vc_data;
-extern unsigned char inverse_translate(struct vc_data *conp, int glyph);
+extern u16 inverse_translate(struct vc_data *conp, int glyph, int use_unicode);
extern unsigned short *set_translate(int m, struct vc_data *vc);
extern int conv_uni_to_pc(struct vc_data *conp, long ucs);
+extern u32 conv_8bit_to_uni(unsigned char c);
#eof
Jan Engelhardt
--
Jan Engelhardt wrote:
> Hi,
>
>
> the following patch had been posted around spring 2005, but has not since
> been included. It is about that hitting the keys <Compose><o><e> for example
> did not produce the utf8 sequence for o-with-diaeresis, but the ASCII/ISO8859-1
> char for o-w-diaeresis, as I seem to recall.
> http://groups.google.de/group/linux.kernel/browse_thread/thread/e60f286969e83d99/55688a5aab326aa7?tvc=2&q=chris+(+%22utf-8%22+OR+%22utf8%22+)+console#55688a5aab326aa7
>
> I am posting an updated version that I hope applies to 2.6.15-rc6.
> Not sure what the correct X-Y-by: procedure is for this, as it originally
> is not my work (see URL).
Looks like I did the same backporting independently. Please look at
http://www.linuxfromscratch.org/~alexander/patches/linux-2.6.14-utf8_input-1.patch
(that does apply to the latest kernels if you use patch -Np1 -l -i
linux-2.6.14-utf8_input-1.patch). Differences between our versions are
described below.
> diff -Ppru linux-2.6.15-rc6-20051219230006/drivers/char/consolemap.c linux-2.6-AS22/drivers/char/consolemap.c
> --- linux-2.6.15-rc6-20051219230006/drivers/char/consolemap.c 2005-12-11 13:42:23.000000000 +0100
> +++ linux-2.6-AS22/drivers/char/consolemap.c 2005-12-19 21:53:25.000000000 +0100
> @@ -178,6 +178,7 @@ struct uni_pagedir {
> unsigned long refcount;
> unsigned long sum;
> unsigned char *inverse_translations[4];
> + u16 *inverse_trans_unicode;
> int readonly;
> };
>
> @@ -208,6 +209,40 @@ static void set_inverse_transl(struct vc
> }
> }
>
> +static void set_inverse_trans_unicode(struct vc_data *conp,
> + struct uni_pagedir *p)
> +{
> + int i, j, k, glyph;
> + u16 **p1, *p2;
> + u16 *q;
> +
I'd insert the "if (!p) return;" statement here, for consistency with
non-Unicode path.
> + q = p->inverse_trans_unicode;
> + if (!q) {
> + q = p->inverse_trans_unicode =
> + kmalloc(MAX_GLYPH * sizeof(u16), GFP_KERNEL);
> + if (!q)
> + return;
> + }
> + memset(q, 0, MAX_GLYPH * sizeof(u16));
> +
> + for (i = 0; i < 32; i++) {
> + p1 = p->uni_pgdir[i];
> + if (!p1)
> + continue;
> + for (j = 0; j < 32; j++) {
> + p2 = p1[j];
> + if (!p2)
> + continue;
> + for (k = 0; k < 64; k++) {
> + glyph = p2[k];
> + if (glyph >= 0 && glyph < MAX_GLYPH
> + && q[glyph] < 32)
> + q[glyph] = (i << 11) + (j << 6) + k;
> + }
> + }
> + }
> +}
> +
> unsigned short *set_translate(int m, struct vc_data *vc)
> {
> inv_translate[vc->vc_num] = m;
> @@ -218,19 +253,29 @@ unsigned short *set_translate(int m, str
> * Inverse translation is impossible for several reasons:
> * 1. The font<->character maps are not 1-1.
> * 2. The text may have been written while a different translation map
> - * was active, or using Unicode.
> + * was active.
> * Still, it is now possible to a certain extent to cut and paste non-ASCII.
> */
> -unsigned char inverse_translate(struct vc_data *conp, int glyph)
> +u16 inverse_translate(struct vc_data *conp, int glyph, int use_unicode)
> {
> struct uni_pagedir *p;
> + int m;
> if (glyph < 0 || glyph >= MAX_GLYPH)
> return 0;
> - else if (!(p = (struct uni_pagedir *)*conp->vc_uni_pagedir_loc) ||
> - !p->inverse_translations[inv_translate[conp->vc_num]])
> + else if (!(p = (struct uni_pagedir *)*conp->vc_uni_pagedir_loc))
> return glyph;
> - else
> - return p->inverse_translations[inv_translate[conp->vc_num]][glyph];
> + else if (use_unicode) {
> + if (!p->inverse_trans_unicode)
> + return glyph;
> + else
> + return p->inverse_trans_unicode[glyph];
> + } else {
> + m = inv_translate[conp->vc_num];
> + if (!p->inverse_translations[m])
> + return glyph;
> + else
> + return p->inverse_translations[m][glyph];
> + }
> }
>
> static void update_user_maps(void)
There is probably a missed set_inverse_trans_unicode(vc_cons[i].d, p);
in update_user_maps() just after set_inverse_transl(vc_cons[i].d, p,
USER_MAP);
> @@ -354,6 +399,10 @@ static void con_release_unimap(struct un
> kfree(p->inverse_translations[i]);
> p->inverse_translations[i] = NULL;
> }
> + if (p->inverse_trans_unicode) {
> + kfree(p->inverse_trans_unicode);
> + p->inverse_trans_unicode = NULL;
> + }
> }
>
> void con_free_unimap(struct vc_data *vc)
> @@ -512,6 +561,7 @@ int con_set_unimap(struct vc_data *vc, u
>
> for (i = 0; i <= 3; i++)
> set_inverse_transl(vc, p, i); /* Update all inverse translations */
> + set_inverse_trans_unicode(vc, p);
>
> return err;
> }
> @@ -562,6 +612,7 @@ int con_set_default_unimap(struct vc_dat
>
> for (i = 0; i <= 3; i++)
> set_inverse_transl(vc, p, i); /* Update all inverse translations */
> + set_inverse_trans_unicode(vc, p);
> dflt = p;
> return err;
> }
> @@ -618,6 +669,19 @@ void con_protect_unimap(struct vc_data *
> p->readonly = rdonly;
> }
>
> +/* may be called during an interrupt */
> +u32 conv_8bit_to_uni(unsigned char c)
> +{
> + /*
> + * Always use USER_MAP. This function is used by the keyboard,
> + * which shouldn't be affected by G0/G1 switching, etc.
> + * If the user map still contains default values, i.e. the
> + * direct-to-font mapping, then assume user is using Latin1.
> + */
> + unsigned short uni = translations[USER_MAP][c];
> + return uni == (0xf000 | c) ? c : uni;
> +}
> +
> int
> conv_uni_to_pc(struct vc_data *conp, long ucs)
> {
> diff -Ppru linux-2.6.15-rc6-20051219230006/drivers/char/keyboard.c linux-2.6-AS22/drivers/char/keyboard.c
> --- linux-2.6.15-rc6-20051219230006/drivers/char/keyboard.c 2005-12-19 21:53:05.000000000 +0100
> +++ linux-2.6-AS22/drivers/char/keyboard.c 2005-12-19 21:53:25.000000000 +0100
> @@ -34,6 +34,7 @@
> #include <linux/init.h>
> #include <linux/slab.h>
>
> +#include <linux/consolemap.h>
> #include <linux/kbd_kern.h>
> #include <linux/kbd_diacr.h>
> #include <linux/vt_kern.h>
> @@ -353,6 +354,15 @@ static void to_utf8(struct vc_data *vc,
My version of to_utf8() takes uint as a second argument and handles
values beyonf 0xffff.
> }
> }
>
> +static void put_8bit(struct vc_data *vc, u8 c)
> +{
> + if (kbd->kbdmode != VC_UNICODE || c < 32 || c == 127)
> + /* Don't translate control chars */
> + put_queue(vc, c);
> + else
> + to_utf8(vc, conv_8bit_to_uni(c));
> +}
> +
> /*
> * Called after returning from RAW mode or when changing consoles - recompute
> * shift_down[] and shift_state from key_down[] maybe called when keymap is
> @@ -413,7 +423,7 @@ static unsigned char handle_diacr(struct
> if (ch == ' ' || ch == d)
> return d;
>
> - put_queue(vc, d);
> + put_8bit(vc, d);
> return ch;
> }
>
> @@ -423,7 +433,7 @@ static unsigned char handle_diacr(struct
> static void fn_enter(struct vc_data *vc, struct pt_regs *regs)
> {
> if (diacr) {
> - put_queue(vc, diacr);
> + put_8bit(vc, diacr);
> diacr = 0;
> }
> put_queue(vc, 13);
> @@ -632,7 +642,7 @@ static void k_self(struct vc_data *vc, u
> diacr = value;
> return;
> }
> - put_queue(vc, value);
> + put_8bit(vc, value);
> }
>
> /*
In k_shift(), there is a difference in the second argument of to_utf8().
My version doesn't have "& 0xffff" because of the abovementioned changes
to this function.
> diff -Ppru linux-2.6.15-rc6-20051219230006/drivers/char/selection.c linux-2.6-AS22/drivers/char/selection.c
> --- linux-2.6.15-rc6-20051219230006/drivers/char/selection.c 2005-12-11 13:42:24.000000000 +0100
> +++ linux-2.6-AS22/drivers/char/selection.c 2005-12-19 21:53:25.000000000 +0100
> @@ -20,6 +20,7 @@
>
> #include <asm/uaccess.h>
>
> +#include <linux/kbd_kern.h>
> #include <linux/vt_kern.h>
> #include <linux/consolemap.h>
> #include <linux/selection.h>
> @@ -34,6 +35,7 @@ extern void poke_blanked_console(void);
> /* Variables for selection control. */
> /* Use a dynamic buffer, instead of static (Dec 1994) */
> struct vc_data *sel_cons; /* must not be disallocated */
> +static int use_unicode;
> static volatile int sel_start = -1; /* cleared by clear_selection */
> static int sel_end;
> static int sel_buffer_lth;
> @@ -54,10 +56,8 @@ static inline void highlight_pointer(con
> complement_pos(sel_cons, where);
> }
>
> -static unsigned char
> -sel_pos(int n)
> -{
> - return inverse_translate(sel_cons, screen_glyph(sel_cons, n));
> +static u16 sel_pos(int n) {
> + return inverse_translate(sel_cons, screen_glyph(sel_cons, n), use_unicode);
> }
>
> /* remove the current selection highlight, if any,
> @@ -86,8 +86,8 @@ static u32 inwordLut[8]={
> 0xFF7FFFFF /* latin-1 accented letters, not division sign */
> };
>
> -static inline int inword(const unsigned char c) {
> - return ( inwordLut[c>>5] >> (c & 0x1F) ) & 1;
> +static inline int inword(const u16 c) {
> + return c > 0xff || ((inwordLut[c >> 5] >> (c & 0x1F)) & 1);
> }
>
> /* set inwordLut contents. Invoked by ioctl(). */
> @@ -108,13 +108,35 @@ static inline unsigned short limit(const
> return (v > u) ? u : v;
> }
>
> +/* stores the char in UTF8 and returns the number of bytes used (1-3) */
> +int store_utf8(u16 c, char *p) {
> + if (c < 0x80) {
> + /* 0******* */
> + p[0] = c;
> + return 1;
> + } else if (c < 0x800) {
> + /* 110***** 10****** */
> + p[0] = 0xc0 | (c >> 6);
> + p[1] = 0x80 | (c & 0x3f);
> + return 2;
> + } else {
> + /* 1110**** 10****** 10****** */
> + p[0] = 0xe0 | (c >> 12);
> + p[1] = 0x80 | ((c >> 6) & 0x3f);
> + p[2] = 0x80 | (c & 0x3f);
> + return 3;
> + }
> +}
> +
> /* set the current selection. Invoked by ioctl() or by kernel code. */
> int set_selection(const struct tiocl_selection __user *sel, struct tty_struct *tty)
> {
> struct vc_data *vc = vc_cons[fg_console].d;
> int sel_mode, new_sel_start, new_sel_end, spc;
> char *bp, *obp;
> - int i, ps, pe;
> + int i, ps, pe, multiplier;
> + u16 c;
> + struct kbd_struct *kbd = kbd_table + fg_console;
>
> poke_blanked_console();
>
> @@ -158,6 +180,7 @@ int set_selection(const struct tiocl_sel
> clear_selection();
> sel_cons = vc_cons[fg_console].d;
> }
> + use_unicode = kbd && kbd->kbdmode == VC_UNICODE;
>
> switch (sel_mode)
> {
> @@ -240,7 +263,8 @@ int set_selection(const struct tiocl_sel
> sel_end = new_sel_end;
>
> /* Allocate a new buffer before freeing the old one ... */
> - bp = kmalloc((sel_end-sel_start)/2+1, GFP_KERNEL);
> + multiplier = use_unicode ? 3 : 1; /* chars can take up to 3 bytes */
> + bp = kmalloc((sel_end-sel_start)/2*multiplier+1, GFP_KERNEL);
> if (!bp) {
> printk(KERN_WARNING "selection: kmalloc() failed\n");
> clear_selection();
> @@ -251,8 +275,12 @@ int set_selection(const struct tiocl_sel
>
> obp = bp;
> for (i = sel_start; i <= sel_end; i += 2) {
> - *bp = sel_pos(i);
> - if (!isspace(*bp++))
> + c = sel_pos(i);
> + if (use_unicode)
> + bp += store_utf8(c, bp);
> + else
> + *bp++ = c;
> + if (!isspace(c))
> obp = bp;
> if (! ((i + 2) % vc->vc_size_row)) {
> /* strip trailing blanks from line and add newline,
> diff -Ppru linux-2.6.15-rc6-20051219230006/include/linux/consolemap.h linux-2.6-AS22/include/linux/consolemap.h
> --- linux-2.6.15-rc6-20051219230006/include/linux/consolemap.h 2005-10-28 02:02:08.000000000 +0200
> +++ linux-2.6-AS22/include/linux/consolemap.h 2005-12-19 21:53:25.000000000 +0100
> @@ -10,6 +10,7 @@
>
> struct vc_data;
>
> -extern unsigned char inverse_translate(struct vc_data *conp, int glyph);
> +extern u16 inverse_translate(struct vc_data *conp, int glyph, int use_unicode);
> extern unsigned short *set_translate(int m, struct vc_data *vc);
> extern int conv_uni_to_pc(struct vc_data *conp, long ucs);
> +extern u32 conv_8bit_to_uni(unsigned char c);
> #eof
>
>
>
> Jan Engelhardt
> Differences between our versions are described below.
What do you think we should do? I have not given this fix any thought,
because it applied fine for long time (+/- fuzz), so I cannot comment on
anything in your version being better or not.
> My version of to_utf8() takes uint as a second argument and handles values
> beyonf 0xffff.
I doubt that there is reason to support characters beyond 0xffff.
CJK is within 0xffff, besides that console fonts just do not have the
capacity to support CJK in a meaningful way.
Jan Engelhardt
--
| Alphagate Systems, http://alphagate.hopto.org/
| jengelh's site, http://jengelh.hopto.org/
Jan Engelhardt wrote:
>>Differences between our versions are described below.
>What do you think we should do? I have not given this fix any thought,
>because it applied fine for long time (+/- fuzz), so I cannot comment on
>anything in your version being better or not.
If you really don't care about characters beyond 0xffff (but see below),
please consider applying this on top of your patch:
--- linux-2.6.15-rc6/drivers/char/consolemap.c 2005-12-25
10:02:11.000000000 +0500
+++ linux-2.6.15-rc6.my/drivers/char/consolemap.c 2005-12-25
10:01:22.000000000 +0500
@@ -216,6 +216,7 @@
u16 **p1, *p2;
u16 *q;
+ if (!p) return;
q = p->inverse_trans_unicode;
if (!q) {
q = p->inverse_trans_unicode =
@@ -289,6 +290,7 @@
p = (struct uni_pagedir *)*vc_cons[i].d->vc_uni_pagedir_loc;
if (p && p != q) {
set_inverse_transl(vc_cons[i].d, p, USER_MAP);
+ set_inverse_trans_unicode(vc_cons[i].d, p);
q = p;
}
}
>>My version of to_utf8() takes uint as a second argument and handles values
>>beyonf 0xffff.
>
>I doubt that there is reason to support characters beyond 0xffff.
>CJK is within 0xffff, besides that console fonts just do not have the
>capacity to support CJK in a meaningful way.
Using console fonts is not the only way to display characters. You
forgot about jfbterm.
--
Alexander E. Patrakov
>> > Differences between our versions are described below.
>
>> What do you think we should do? I have not given this fix any thought,
>> because it applied fine for long time (+/- fuzz), so I cannot comment on
>> anything in your version being better or not.
>
> If you really don't care about characters beyond 0xffff (but see below),
> please consider applying this on top of your patch:
Could you just send the whole thing to mainline once 2.6.16 is out? Thanks.
Jan Engelhardt
--
Hi Alex,
>> Yes, please assume this line on all utf8 composing patches I submitted to LKML:
>> But this line already exists in http://lkml.org/lkml/2006/3/20/571 :)
>Must have missed it.
>
Your patch will not apply 100% cleanly to 2.6.17-rc3, and in fact it
seems like it is not needed anymore (composing works without it again).
Can you confirm that?
Jan Engelhardt
--
Jan Engelhardt wrote:
> Hi Alex,
>
>
>>> Yes, please assume this line on all utf8 composing patches I submitted to LKML:
>>> But this line already exists in http://lkml.org/lkml/2006/3/20/571 :)
>>>
>> Must have missed it.
>>
>>
>
> Your patch will not apply 100% cleanly to 2.6.17-rc3, and in fact it
> seems like it is not needed anymore (composing works without it again).
> Can you confirm that?
>
>
>
> Jan Engelhardt
>
No, I can't.
What they did is to inline the equivalent of the put_8bit() function,
minus the call to conv_8bit_to_uni(). So, the result is that the result
of the composing operation is still taken from the accent_table, and
thus cannot be more than "unsigned char" allows (but is correctly
converted to UTF-8). See the definition of struct kbdiacr in
include/linux/kd.h. So, the result of such operation can be only in the
Latin-1 range of Unicode. This is not acceptable for languages that use
dead keys and characters from Latin-2 set.
In short: in UTF-8 mode, it is possible to generate characters from
Latin-1 set by composing, and they are generated correctly. It is
impossible to generate characters outside of Latin-1 (unlike the
situation with my patch).
Both the current situation and my patch share the defect that an accent
cannot be put on top of a multibyte character, such as Greek letter alpha.
The updated patch is pasted below. It works and contains the needed
functionality (plus Unicode copy-and-paste functionality), but is
ideologically wrong (it would be more correct to introduce a new version
of struct kbdiacr and new version of kbd). However, it is currently the
only known way to generate characters outside of Latin-1 by composing in
UTF-8 mode.
Submitted by: Alexander E. Patrakov <[email protected]>
Signed-off-by: Alexander E. Patrakov <[email protected]>
Date: 2006-05-09
Initial Package Version: 2.6.17-rc3
Upstream Status: Discussion on LKML
Origin: http://chris.heathens.co.nz/linux/downloads/patches-2.6.4-cdh1.tar.gz
Porting to linux-2.6.17-rc3 by Alexander E. Patrakov
Description: This patch fixes dead keys and copy/paste of non-ASCII characters
in UTF-8 mode on Linux console.
See more details about the original patch at:
http://chris.heathens.co.nz/linux/utf8.html
---
diff -ur linux-2.6.17-rc3.orig/drivers/char/consolemap.c linux-2.6.17-rc3.my/drivers/char/consolemap.c
--- linux-2.6.17-rc3.orig/drivers/char/consolemap.c 2005-12-25 10:00:12.000000000 +0500
+++ linux-2.6.17-rc3.my/drivers/char/consolemap.c 2005-12-25 10:01:22.000000000 +0500
@@ -178,6 +178,7 @@
unsigned long refcount;
unsigned long sum;
unsigned char *inverse_translations[4];
+ u16 *inverse_trans_unicode;
int readonly;
};
@@ -208,6 +209,41 @@
}
}
+static void set_inverse_trans_unicode(struct vc_data *conp,
+ struct uni_pagedir *p)
+{
+ int i, j, k, glyph;
+ u16 **p1, *p2;
+ u16 *q;
+
+ if (!p) return;
+ q = p->inverse_trans_unicode;
+ if (!q) {
+ q = p->inverse_trans_unicode =
+ kmalloc(MAX_GLYPH * sizeof(u16), GFP_KERNEL);
+ if (!q)
+ return;
+ }
+ memset(q, 0, MAX_GLYPH * sizeof(u16));
+
+ for (i = 0; i < 32; i++) {
+ p1 = p->uni_pgdir[i];
+ if (!p1)
+ continue;
+ for (j = 0; j < 32; j++) {
+ p2 = p1[j];
+ if (!p2)
+ continue;
+ for (k = 0; k < 64; k++) {
+ glyph = p2[k];
+ if (glyph >= 0 && glyph < MAX_GLYPH
+ && q[glyph] < 32)
+ q[glyph] = (i << 11) + (j << 6) + k;
+ }
+ }
+ }
+}
+
unsigned short *set_translate(int m, struct vc_data *vc)
{
inv_translate[vc->vc_num] = m;
@@ -218,19 +254,29 @@
* Inverse translation is impossible for several reasons:
* 1. The font<->character maps are not 1-1.
* 2. The text may have been written while a different translation map
- * was active, or using Unicode.
+ * was active.
* Still, it is now possible to a certain extent to cut and paste non-ASCII.
*/
-unsigned char inverse_translate(struct vc_data *conp, int glyph)
+u16 inverse_translate(struct vc_data *conp, int glyph, int use_unicode)
{
struct uni_pagedir *p;
+ int m;
if (glyph < 0 || glyph >= MAX_GLYPH)
return 0;
- else if (!(p = (struct uni_pagedir *)*conp->vc_uni_pagedir_loc) ||
- !p->inverse_translations[inv_translate[conp->vc_num]])
+ else if (!(p = (struct uni_pagedir *)*conp->vc_uni_pagedir_loc))
return glyph;
- else
- return p->inverse_translations[inv_translate[conp->vc_num]][glyph];
+ else if (use_unicode) {
+ if (!p->inverse_trans_unicode)
+ return glyph;
+ else
+ return p->inverse_trans_unicode[glyph];
+ } else {
+ m = inv_translate[conp->vc_num];
+ if (!p->inverse_translations[m])
+ return glyph;
+ else
+ return p->inverse_translations[m][glyph];
+ }
}
static void update_user_maps(void)
@@ -244,6 +290,7 @@
p = (struct uni_pagedir *)*vc_cons[i].d->vc_uni_pagedir_loc;
if (p && p != q) {
set_inverse_transl(vc_cons[i].d, p, USER_MAP);
+ set_inverse_trans_unicode(vc_cons[i].d, p);
q = p;
}
}
@@ -354,6 +401,10 @@
kfree(p->inverse_translations[i]);
p->inverse_translations[i] = NULL;
}
+ if (p->inverse_trans_unicode) {
+ kfree(p->inverse_trans_unicode);
+ p->inverse_trans_unicode = NULL;
+ }
}
void con_free_unimap(struct vc_data *vc)
@@ -512,6 +563,7 @@
for (i = 0; i <= 3; i++)
set_inverse_transl(vc, p, i); /* Update all inverse translations */
+ set_inverse_trans_unicode(vc, p);
return err;
}
@@ -562,6 +614,7 @@
for (i = 0; i <= 3; i++)
set_inverse_transl(vc, p, i); /* Update all inverse translations */
+ set_inverse_trans_unicode(vc, p);
dflt = p;
return err;
}
@@ -618,6 +671,19 @@
p->readonly = rdonly;
}
+/* may be called during an interrupt */
+u32 conv_8bit_to_uni(unsigned char c)
+{
+ /*
+ * Always use USER_MAP. This function is used by the keyboard,
+ * which shouldn't be affected by G0/G1 switching, etc.
+ * If the user map still contains default values, i.e. the
+ * direct-to-font mapping, then assume user is using Latin1.
+ */
+ unsigned short uni = translations[USER_MAP][c];
+ return uni == (0xf000 | c) ? c : uni;
+}
+
int
conv_uni_to_pc(struct vc_data *conp, long ucs)
{
diff -ur linux-2.6.17-rc3.orig/drivers/char/keyboard.c linux-2.6.17-rc3.my/drivers/char/keyboard.c
--- linux-2.6.17-rc3.orig/drivers/char/keyboard.c 2006-05-09 02:39:23.000000000 +0600
+++ linux-2.6.17-rc3.my/drivers/char/keyboard.c 2006-05-09 06:38:24.000000000 +0600
@@ -34,6 +34,7 @@
#include <linux/init.h>
#include <linux/slab.h>
+#include <linux/consolemap.h>
#include <linux/kbd_kern.h>
#include <linux/kbd_diacr.h>
#include <linux/vt_kern.h>
@@ -329,10 +330,9 @@
* Many other routines do put_queue, but I think either
* they produce ASCII, or they produce some user-assigned
* string, and in both cases we might assume that it is
- * in utf-8 already. UTF-8 is defined for words of up to 31 bits,
- * but we need only 16 bits here
+ * in utf-8 already.
*/
-static void to_utf8(struct vc_data *vc, ushort c)
+static void to_utf8(struct vc_data *vc, uint c)
{
if (c < 0x80)
/* 0******* */
@@ -341,11 +341,21 @@
/* 110***** 10****** */
put_queue(vc, 0xc0 | (c >> 6));
put_queue(vc, 0x80 | (c & 0x3f));
- } else {
+ } else if (c < 0x10000) {
+ if (c >= 0xD800 && c < 0xE000)
+ return;
+ if (c == 0xFFFF)
+ return;
/* 1110**** 10****** 10****** */
put_queue(vc, 0xe0 | (c >> 12));
put_queue(vc, 0x80 | ((c >> 6) & 0x3f));
put_queue(vc, 0x80 | (c & 0x3f));
+ } else if (c < 0x110000) {
+ /* 11110*** 10****** 10****** 10****** */
+ put_queue(vc, 0xf0 | (c >> 18));
+ put_queue(vc, 0x80 | ((c >> 12) & 0x3f));
+ put_queue(vc, 0x80 | ((c >> 6) & 0x3f));
+ put_queue(vc, 0x80 | (c & 0x3f));
}
}
@@ -414,7 +424,7 @@
return d;
if (kbd->kbdmode == VC_UNICODE)
- to_utf8(vc, d);
+ to_utf8(vc, conv_8bit_to_uni(d));
else if (d < 0x100)
put_queue(vc, d);
@@ -428,7 +438,7 @@
{
if (diacr) {
if (kbd->kbdmode == VC_UNICODE)
- to_utf8(vc, diacr);
+ to_utf8(vc, conv_8bit_to_uni(diacr));
else if (diacr < 0x100)
put_queue(vc, diacr);
diacr = 0;
@@ -640,7 +650,7 @@
return;
}
if (kbd->kbdmode == VC_UNICODE)
- to_utf8(vc, value);
+ to_utf8(vc, conv_8bit_to_uni(value));
else if (value < 0x100)
put_queue(vc, value);
}
@@ -798,7 +808,7 @@
/* kludge */
if (up_flag && shift_state != old_state && npadch != -1) {
if (kbd->kbdmode == VC_UNICODE)
- to_utf8(vc, npadch & 0xffff);
+ to_utf8(vc, npadch);
else
put_queue(vc, npadch & 0xff);
npadch = -1;
diff -ur linux-2.6.17-rc3.orig/drivers/char/selection.c linux-2.6.17-rc3.my/drivers/char/selection.c
--- linux-2.6.17-rc3.orig/drivers/char/selection.c 2005-12-25 10:00:12.000000000 +0500
+++ linux-2.6.17-rc3.my/drivers/char/selection.c 2005-12-25 10:01:22.000000000 +0500
@@ -20,6 +20,7 @@
#include <asm/uaccess.h>
+#include <linux/kbd_kern.h>
#include <linux/vt_kern.h>
#include <linux/consolemap.h>
#include <linux/selection.h>
@@ -34,6 +35,7 @@
/* Variables for selection control. */
/* Use a dynamic buffer, instead of static (Dec 1994) */
struct vc_data *sel_cons; /* must not be disallocated */
+static int use_unicode;
static volatile int sel_start = -1; /* cleared by clear_selection */
static int sel_end;
static int sel_buffer_lth;
@@ -54,10 +56,11 @@
complement_pos(sel_cons, where);
}
-static unsigned char
+static u16
sel_pos(int n)
{
- return inverse_translate(sel_cons, screen_glyph(sel_cons, n));
+ return inverse_translate(sel_cons, screen_glyph(sel_cons, n),
+ use_unicode);
}
/* remove the current selection highlight, if any,
@@ -86,8 +89,8 @@
0xFF7FFFFF /* latin-1 accented letters, not division sign */
};
-static inline int inword(const unsigned char c) {
- return ( inwordLut[c>>5] >> (c & 0x1F) ) & 1;
+static inline int inword(const u16 c) {
+ return c > 0xff || (( inwordLut[c>>5] >> (c & 0x1F) ) & 1);
}
/* set inwordLut contents. Invoked by ioctl(). */
@@ -108,13 +111,36 @@
return (v > u) ? u : v;
}
+/* stores the char in UTF8 and returns the number of bytes used (1-3) */
+int store_utf8(u16 c, char *p)
+{
+ if (c < 0x80) {
+ /* 0******* */
+ p[0] = c;
+ return 1;
+ } else if (c < 0x800) {
+ /* 110***** 10****** */
+ p[0] = 0xc0 | (c >> 6);
+ p[1] = 0x80 | (c & 0x3f);
+ return 2;
+ } else {
+ /* 1110**** 10****** 10****** */
+ p[0] = 0xe0 | (c >> 12);
+ p[1] = 0x80 | ((c >> 6) & 0x3f);
+ p[2] = 0x80 | (c & 0x3f);
+ return 3;
+ }
+}
+
/* set the current selection. Invoked by ioctl() or by kernel code. */
int set_selection(const struct tiocl_selection __user *sel, struct tty_struct *tty)
{
struct vc_data *vc = vc_cons[fg_console].d;
int sel_mode, new_sel_start, new_sel_end, spc;
char *bp, *obp;
- int i, ps, pe;
+ int i, ps, pe, multiplier;
+ u16 c;
+ struct kbd_struct *kbd = kbd_table + fg_console;
poke_blanked_console();
@@ -158,7 +184,8 @@
clear_selection();
sel_cons = vc_cons[fg_console].d;
}
-
+ use_unicode = kbd && kbd->kbdmode == VC_UNICODE;
+
switch (sel_mode)
{
case TIOCL_SELCHAR: /* character-by-character selection */
@@ -240,7 +267,8 @@
sel_end = new_sel_end;
/* Allocate a new buffer before freeing the old one ... */
- bp = kmalloc((sel_end-sel_start)/2+1, GFP_KERNEL);
+ multiplier = use_unicode ? 3 : 1; /* chars can take up to 3 bytes */
+ bp = kmalloc((sel_end-sel_start)/2*multiplier+1, GFP_KERNEL);
if (!bp) {
printk(KERN_WARNING "selection: kmalloc() failed\n");
clear_selection();
@@ -251,8 +279,12 @@
obp = bp;
for (i = sel_start; i <= sel_end; i += 2) {
- *bp = sel_pos(i);
- if (!isspace(*bp++))
+ c = sel_pos(i);
+ if (use_unicode)
+ bp += store_utf8(c, bp);
+ else
+ *bp++ = c;
+ if (!isspace(c))
obp = bp;
if (! ((i + 2) % vc->vc_size_row)) {
/* strip trailing blanks from line and add newline,
diff -ur linux-2.6.17-rc3.orig/include/linux/consolemap.h linux-2.6.17-rc3.my/include/linux/consolemap.h
--- linux-2.6.17-rc3.orig/include/linux/consolemap.h 2005-12-25 10:00:13.000000000 +0500
+++ linux-2.6.17-rc3.my/include/linux/consolemap.h 2005-12-25 10:01:22.000000000 +0500
@@ -10,6 +10,7 @@
struct vc_data;
-extern unsigned char inverse_translate(struct vc_data *conp, int glyph);
+extern u16 inverse_translate(struct vc_data *conp, int glyph, int use_unicode);
extern unsigned short *set_translate(int m, struct vc_data *vc);
extern int conv_uni_to_pc(struct vc_data *conp, long ucs);
+extern u32 conv_8bit_to_uni(unsigned char c);
--
Alexander E. Patrakov
Hi Alex,
Just for the archive...
On Tuesday, 9. May 2006 09:49, Alexander E. Patrakov wrote:
> Both the current situation and my patch share the defect that an accent
> cannot be put on top of a multibyte character, such as Greek letter alpha.
This is also a problem for proper Vietnamese console support,
which requires two accents per character.
Like "~" on top of "^" ontop of "a".
Regards
Ingo Oeser
>Just for the archive...
>
>On Tuesday, 9. May 2006 09:49, Alexander E. Patrakov wrote:
>> Both the current situation and my patch share the defect that an accent
>> cannot be put on top of a multibyte character, such as Greek letter alpha.
With 80x25, that [almost] would not be possible either because of the 256
character limit. Of course the concern is totally valid for fbterm.
Jan Engelhardt
--
Jan Engelhardt wrote:
>> Just for the archive...
>>
>> On Tuesday, 9. May 2006 09:49, Alexander E. Patrakov wrote:
>>> Both the current situation and my patch share the defect that an accent
>>> cannot be put on top of a multibyte character, such as Greek letter alpha.
>
> With 80x25, that [almost] would not be possible either because of the 256
> character limit. Of course the concern is totally valid for fbterm.
This concern is totally valid for the 80x25 console and the standard "gr" keymap:
compose '\'' 'α' to 'ά'
and the ά character is indeed present in the iso07.16 font and used in *.po
files from console-tools.
--
Alexander E. Patrakov