diff options
author | jan.nijtmans <nijtmans@users.sourceforge.net> | 2019-12-13 15:09:59 (GMT) |
---|---|---|
committer | jan.nijtmans <nijtmans@users.sourceforge.net> | 2019-12-13 15:09:59 (GMT) |
commit | 9118bcfb5fc11ee3cb496e1da2d648c6ea447a6b (patch) | |
tree | 964c8e51f9e9947f32d8a5cbe0d75c29a32b71df /generic/tclTomMath.decls | |
parent | 8f533db35fcc6fecc814d6e6e5b966fb7225c045 (diff) | |
download | tcl-9118bcfb5fc11ee3cb496e1da2d648c6ea447a6b.zip tcl-9118bcfb5fc11ee3cb496e1da2d648c6ea447a6b.tar.gz tcl-9118bcfb5fc11ee3cb496e1da2d648c6ea447a6b.tar.bz2 |
More tweaks to libtommath functions signatures: No need any more to stay binary compatible with Tcl 8.x, so we can use the mp_digit type now.
Diffstat (limited to 'generic/tclTomMath.decls')
-rw-r--r-- | generic/tclTomMath.decls | 25 |
1 files changed, 13 insertions, 12 deletions
diff --git a/generic/tclTomMath.decls b/generic/tclTomMath.decls index 9afb284..a47f7ef 100644 --- a/generic/tclTomMath.decls +++ b/generic/tclTomMath.decls @@ -33,7 +33,7 @@ declare 2 { mp_err MP_WUR TclBN_mp_add(const mp_int *a, const mp_int *b, mp_int *c) } declare 3 { - mp_err MP_WUR TclBN_mp_add_d(const mp_int *a, unsigned int b, mp_int *c) + mp_err MP_WUR TclBN_mp_add_d(const mp_int *a, mp_digit b, mp_int *c) } declare 4 { mp_err MP_WUR TclBN_mp_and(const mp_int *a, const mp_int *b, mp_int *c) @@ -51,7 +51,7 @@ declare 8 { mp_ord MP_WUR TclBN_mp_cmp(const mp_int *a, const mp_int *b) } declare 9 { - mp_ord MP_WUR TclBN_mp_cmp_d(const mp_int *a, unsigned int b) + mp_ord MP_WUR TclBN_mp_cmp_d(const mp_int *a, mp_digit b) } declare 10 { mp_ord MP_WUR TclBN_mp_cmp_mag(const mp_int *a, const mp_int *b) @@ -66,7 +66,7 @@ declare 13 { mp_err MP_WUR TclBN_mp_div(const mp_int *a, const mp_int *b, mp_int *q, mp_int *r) } declare 14 { - mp_err MP_WUR TclBN_mp_div_d(const mp_int *a, unsigned int b, mp_int *q, unsigned int *r) + mp_err MP_WUR TclBN_mp_div_d(const mp_int *a, mp_digit b, mp_int *q, mp_digit *r) } declare 15 { mp_err MP_WUR TclBN_mp_div_2(const mp_int *a, mp_int *q) @@ -76,13 +76,13 @@ declare 16 { } # Removed in 9.0 #declare 17 {deprecated {is private function in libtommath}} { -# mp_err TclBN_mp_div_3(const mp_int *a, mp_int *q, unsigned int *r) +# mp_err TclBN_mp_div_3(const mp_int *a, mp_int *q, mp_digit *r) #} declare 18 { void TclBN_mp_exch(mp_int *a, mp_int *b) } declare 19 { - mp_err MP_WUR TclBN_mp_expt_u32(const mp_int *a, unsigned int b, mp_int *c) + mp_err MP_WUR TclBN_mp_expt_u32(const mp_int *a, uint32_t b, mp_int *c) } declare 20 { mp_err MP_WUR TclBN_mp_grow(mp_int *a, int size) @@ -97,7 +97,7 @@ declare 23 { mp_err MP_WUR TclBN_mp_init_multi(mp_int *a, ...) } declare 24 { - mp_err MP_WUR TclBN_mp_init_set(mp_int *a, unsigned int b) + mp_err MP_WUR TclBN_mp_init_set(mp_int *a, mp_digit b) } declare 25 { mp_err MP_WUR TclBN_mp_init_size(mp_int *a, int size) @@ -115,7 +115,7 @@ declare 29 { mp_err MP_WUR TclBN_mp_mul(const mp_int *a, const mp_int *b, mp_int *p) } declare 30 { - mp_err MP_WUR TclBN_mp_mul_d(const mp_int *a, unsigned int b, mp_int *p) + mp_err MP_WUR TclBN_mp_mul_d(const mp_int *a, mp_digit b, mp_int *p) } declare 31 { mp_err MP_WUR TclBN_mp_mul_2(const mp_int *a, mp_int *p) @@ -156,7 +156,7 @@ declare 42 { mp_err MP_WUR TclBN_mp_sub(const mp_int *a, const mp_int *b, mp_int *c) } declare 43 { - mp_err MP_WUR TclBN_mp_sub_d(const mp_int *a, unsigned int b, mp_int *c) + mp_err MP_WUR TclBN_mp_sub_d(const mp_int *a, mp_digit b, mp_int *c) } # Removed in 9.0 #declare 44 { @@ -204,7 +204,7 @@ declare 66 { # Removed in 9.0 #declare 67 { -# mp_err TclBN_mp_expt_d_ex(const mp_int *a, unsigned int b, mp_int *c, int fast) +# mp_err TclBN_mp_expt_d_ex(const mp_int *a, mp_digit b, mp_int *c, int fast) #} # Added in libtommath 1.0.1 declare 68 { @@ -238,9 +238,10 @@ declare 76 { declare 78 { int MP_WUR TclBN_mp_to_ubin(const mp_int *a, unsigned char *buf, size_t maxlen, size_t *written) } -declare 79 { - mp_err MP_WUR TclBN_mp_div_ld(const mp_int *a, uint64_t b, mp_int *q, uint64_t *r) -} +# Removed in 9.0 +#declare 79 { +# mp_err MP_WUR TclBN_mp_div_ld(const mp_int *a, mp_digit b, mp_int *q, mp_digit *r) +#} declare 80 { int MP_WUR TclBN_mp_to_radix(const mp_int *a, char *str, size_t maxlen, size_t *written, int radix) } |