aboutsummaryrefslogtreecommitdiff
path: root/src/lib/extfmt.rs
diff options
context:
space:
mode:
Diffstat (limited to 'src/lib/extfmt.rs')
-rw-r--r--src/lib/extfmt.rs258
1 files changed, 256 insertions, 2 deletions
diff --git a/src/lib/extfmt.rs b/src/lib/extfmt.rs
index c5691fe9..93d4d7de 100644
--- a/src/lib/extfmt.rs
+++ b/src/lib/extfmt.rs
@@ -28,7 +28,7 @@ import option::some;
*/
// Functions used by the fmt extension at compile time
-mod CT {
+mod ct {
tag signedness {
signed;
unsigned;
@@ -305,8 +305,262 @@ mod CT {
// decisions made a runtime. If it proves worthwhile then some of these
// conditions can be evaluated at compile-time. For now though it's cleaner to
// implement it this way, I think.
-mod RT {
+mod rt {
+
+ tag flag {
+ flag_left_justify;
+ flag_left_zero_pad;
+ flag_space_for_sign;
+ flag_sign_always;
+ flag_alternate;
+ // FIXME: This is a hack to avoid creating 0-length vec exprs,
+ // which have some difficulty typechecking currently. See
+ // comments in front::extfmt::make_flags
+ flag_none;
+ }
+
+ tag count {
+ count_is(int);
+ count_implied;
+ }
+
+ tag ty {
+ ty_default;
+ ty_bits;
+ ty_hex_upper;
+ ty_hex_lower;
+ ty_octal;
+ }
+
+ // FIXME: May not want to use a vector here for flags;
+ // instead just use a bool per flag
+ type conv = rec(vec[flag] flags,
+ count width,
+ count precision,
+ ty ty);
+
+ fn conv_int(&conv cv, int i) -> str {
+ auto radix = 10u;
+ auto prec = get_int_precision(cv);
+ auto s = int_to_str_prec(i, radix, prec);
+ if (0 <= i) {
+ if (have_flag(cv.flags, flag_sign_always)) {
+ s = "+" + s;
+ } else if (have_flag(cv.flags, flag_space_for_sign)) {
+ s = " " + s;
+ }
+ }
+ ret pad(cv, s, pad_signed);
+ }
+ fn conv_uint(&conv cv, uint u) -> str {
+ auto prec = get_int_precision(cv);
+ auto res;
+ alt (cv.ty) {
+ case (ty_default) {
+ res = uint_to_str_prec(u, 10u, prec);
+ }
+ case (ty_hex_lower) {
+ res = uint_to_str_prec(u, 16u, prec);
+ }
+ case (ty_hex_upper) {
+ res = _str::to_upper(uint_to_str_prec(u, 16u, prec));
+ }
+ case (ty_bits) {
+ res = uint_to_str_prec(u, 2u, prec);
+ }
+ case (ty_octal) {
+ res = uint_to_str_prec(u, 8u, prec);
+ }
+ }
+ ret pad(cv, res, pad_unsigned);
+ }
+
+ fn conv_bool(&conv cv, bool b) -> str {
+ auto s;
+ if (b) {
+ s = "true";
+ } else {
+ s = "false";
+ }
+ // run the boolean conversion through the string conversion logic,
+ // giving it the same rules for precision, etc.
+ ret conv_str(cv, s);
+ }
+
+ fn conv_char(&conv cv, char c) -> str {
+ ret pad(cv, _str::from_char(c), pad_nozero);
+ }
+
+ fn conv_str(&conv cv, str s) -> str {
+ auto unpadded = s;
+ alt (cv.precision) {
+ case (count_implied) {
+ }
+ case (count_is(?max)) {
+ // For strings, precision is the maximum characters displayed
+ if (max as uint < _str::char_len(s)) {
+ // FIXME: substr works on bytes, not chars!
+ unpadded = _str::substr(s, 0u, max as uint);
+ }
+ }
+ }
+ ret pad(cv, unpadded, pad_nozero);
+ }
+
+ // Convert an int to string with minimum number of digits. If precision is
+ // 0 and num is 0 then the result is the empty string.
+ fn int_to_str_prec(int num, uint radix, uint prec) -> str {
+ if (num < 0) {
+ ret "-" + uint_to_str_prec((-num) as uint, radix, prec);
+ } else {
+ ret uint_to_str_prec(num as uint, radix, prec);
+ }
+ }
+
+ // Convert a uint to string with a minimum number of digits. If precision
+ // is 0 and num is 0 then the result is the empty string. Could move this
+ // to _uint: but it doesn't seem all that useful.
+ fn uint_to_str_prec(uint num, uint radix, uint prec) -> str {
+ auto s;
+
+ if (prec == 0u && num == 0u) {
+ s = "";
+ } else {
+ s = _uint::to_str(num, radix);
+ auto len = _str::char_len(s);
+ if (len < prec) {
+ auto diff = prec - len;
+ auto pad = str_init_elt('0', diff);
+ s = pad + s;
+ }
+ }
+
+ ret s;
+ }
+
+ fn get_int_precision(&conv cv) -> uint {
+ alt (cv.precision) {
+ case (count_is(?c)) {
+ ret c as uint;
+ }
+ case (count_implied) {
+ ret 1u;
+ }
+ }
+ }
+
+ // FIXME: This might be useful in _str: but needs to be utf8 safe first
+ fn str_init_elt(char c, uint n_elts) -> str {
+ auto svec = _vec::init_elt[u8](c as u8, n_elts);
+ // FIXME: Using unsafe_from_bytes because rustboot
+ // can't figure out the is_utf8 predicate on from_bytes?
+ ret _str::unsafe_from_bytes(svec);
+ }
+
+ tag pad_mode {
+ pad_signed;
+ pad_unsigned;
+ pad_nozero;
+ }
+
+ fn pad(&conv cv, str s, pad_mode mode) -> str {
+ auto uwidth;
+ alt (cv.width) {
+ case (count_implied) {
+ ret s;
+ }
+ case (count_is(?width)) {
+ // FIXME: Maybe width should be uint
+ uwidth = width as uint;
+ }
+ }
+
+ auto strlen = _str::char_len(s);
+ if (uwidth <= strlen) {
+ ret s;
+ }
+
+ auto padchar = ' ';
+ auto diff = uwidth - strlen;
+ if (have_flag(cv.flags, flag_left_justify)) {
+ auto padstr = str_init_elt(padchar, diff);
+ ret s + padstr;
+ }
+
+ auto might_zero_pad = false;
+ auto signed = false;
+
+ alt (mode) {
+ case (pad_nozero) {
+ // fallthrough
+ }
+ case (pad_signed) {
+ might_zero_pad = true;
+ signed = true;
+ }
+ case (pad_unsigned) {
+ might_zero_pad = true;
+ }
+ }
+
+ fn have_precision(&conv cv) -> bool {
+ alt (cv.precision) {
+ case (count_implied) {
+ ret false;
+ }
+ case (_) {
+ ret true;
+ }
+ }
+ }
+
+ auto zero_padding = false;
+ if (might_zero_pad
+ && have_flag(cv.flags, flag_left_zero_pad)
+ && !have_precision(cv)) {
+
+ padchar = '0';
+ zero_padding = true;
+ }
+
+ auto padstr = str_init_elt(padchar, diff);
+
+ // This is completely heinous. If we have a signed value then
+ // potentially rip apart the intermediate result and insert some
+ // zeros. It may make sense to convert zero padding to a precision
+ // instead.
+ if (signed
+ && zero_padding
+ && _str::byte_len(s) > 0u) {
+
+ auto head = s.(0);
+ if (head == '+' as u8
+ || head == '-' as u8
+ || head == ' ' as u8) {
+
+ auto headstr = _str::unsafe_from_bytes(vec(head));
+ auto bytelen = _str::byte_len(s);
+ auto numpart = _str::substr(s, 1u, bytelen - 1u);
+ ret headstr + padstr + numpart;
+ }
+ }
+ ret padstr + s;
+ }
+
+ fn have_flag(vec[flag] flags, flag f) -> bool {
+ for (flag candidate in flags) {
+ if (candidate == f) {
+ ret true;
+ }
+ }
+ ret false;
+ }
+}
+
+// FIXME: This is a temporary duplication of the rt mod that only
+// needs to exist to accomodate the stage0 compiler until the next snapshot
+mod RT {
tag flag {
flag_left_justify;
flag_left_zero_pad;