[PATCH WIP/RFC 11/11] gas: have scrubber retain more whitespace
Jan Beulich
jbeulich@suse.com
Fri Jun 28 13:31:20 GMT 2024
According to the description os the state machine, the expectation
appears to be that (leaving aside labels) any insn mnemonic or
directive would be followed by a comma separated list of operands. That
may have been true very long ago, but the latest with the advent of more
elaborate macros this isn't rhe case anymore. Neither macro parameters
in macro definitions nor macro arguments in macro invocations are
required to be separated by commas. Hence whitespace serves a crucial
role there. Plus even without "real" macros issues exist, in e.g.
.irp n, ...
insn\n\(suffix) operand1, operand2
.endr
Whitespace following the closing parenthesis would have been removed
(ahead of even processing the .irp), as the "opcode" was deemed to have
ended earlier already.
Therefore, squash the distinction between "opcode" and operands, i.e.
fold state 10 back into state 3. Also drop most of the distinction
between "symbol chars" and "relatively normal" ones. Not entirely
unexpectedly this results in the need to skip whitespace in a few more
places in arch-specific code.
---
The comma related special casing could of course be extended to adjacent
two-char comments; I simply didn't deem this as important.
WIP: likely there's also fallout for further architectures
WIP: see about dropping the D10V special case (not helped by there not
being a dedicated maintainer)
WIP: revisit the TIC6X special case
WIP: testcase(s)
WIP: NEWS entry (perhaps also to warn people of posible perceived
regressions, along the lines of what "Arm: correct macro use in gas
testsuite" addresses)
--- a/gas/app.c
+++ b/gas/app.c
@@ -423,16 +423,18 @@ do_scrub_chars (size_t (*get) (char *, s
/*State 0: beginning of normal line
1: After first whitespace on line (flush more white)
- 2: After first non-white (opcode) on line (keep 1white)
- 3: after second white on line (into operands) (flush white)
+ 2: After first non-white (opcode or maybe label when they're followed
+ by colons) on line (keep 1white)
+ 3: after subsequent white on line (typically into operands)
+ (flush more white)
4: after putting out a .linefile, put out digits
5: parsing a string, then go to old-state
6: putting out \ escape in a "d string.
7: no longer used
8: no longer used
- 9: After seeing symbol char in state 3 (keep 1white after symchar)
- 10: After seeing whitespace in state 9 (keep white before symchar)
- 11: After seeing a symbol character in state 0 (eg a label definition)
+ 9: After seeing non-white in state 3 (keep 1white)
+ 10: no longer used
+ 11: After seeing a non-white character in state 0 (eg a label definition)
-1: output string in out_string and go to the state in old_state
12: no longer used
#ifdef DOUBLEBAR_PARALLEL
@@ -899,7 +901,11 @@ do_scrub_chars (size_t (*get) (char *, s
&& (state < 1 || strchr (tc_comment_chars, ch)))
|| IS_NEWLINE (ch)
|| IS_LINE_SEPARATOR (ch)
- || IS_PARALLEL_SEPARATOR (ch))
+ || IS_PARALLEL_SEPARATOR (ch)
+ /* See comma related comment near the bottom of the function.
+ Reasoning equally applies to whitespace preceding a comma in
+ most cases. */
+ || (ch == ',' && state > 2))
{
if (scrub_m68k_mri)
{
@@ -942,6 +948,7 @@ do_scrub_chars (size_t (*get) (char *, s
character at the beginning of a line. */
goto recycle;
case 2:
+ case 9:
state = 3;
if (to + 1 < toend)
{
@@ -965,20 +972,6 @@ do_scrub_chars (size_t (*get) (char *, s
break;
}
goto recycle; /* Sp in operands */
- case 9:
- case 10:
-#ifndef TC_KEEP_OPERAND_SPACES
- if (scrub_m68k_mri)
-#endif
- {
- /* In MRI mode, we keep these spaces. */
- state = 3;
- UNGET (ch);
- PUT (' ');
- break;
- }
- state = 10; /* Sp after symbol char */
- goto recycle;
case 11:
if (LABELS_WITHOUT_COLONS || flag_m68k_mri)
state = 1;
@@ -1049,27 +1042,17 @@ do_scrub_chars (size_t (*get) (char *, s
{
if (ch2 != EOF)
UNGET (ch2);
- if (state == 9 || state == 10)
- state = 3;
+ if (state == 1)
+ state = 2;
+ else if (state == 3)
+ state = 9;
PUT (ch);
}
break;
case LEX_IS_STRINGQUOTE:
quotechar = ch;
- if (state == 10)
- {
- /* Preserve the whitespace in foo "bar". */
- UNGET (ch);
- state = 3;
- PUT (' ');
-
- /* PUT didn't jump out. We could just break, but we
- know what will happen, so optimize a bit. */
- ch = GET ();
- old_state = 9;
- }
- else if (state == 3)
+ if (state == 3)
old_state = 9;
else
old_state = state;
@@ -1088,14 +1071,6 @@ do_scrub_chars (size_t (*get) (char *, s
UNGET (c);
}
#endif
- if (state == 10)
- {
- /* Preserve the whitespace in foo 'b'. */
- UNGET (ch);
- state = 3;
- PUT (' ');
- break;
- }
ch = GET ();
if (ch == EOF)
{
@@ -1130,10 +1105,7 @@ do_scrub_chars (size_t (*get) (char *, s
PUT (out_buf[0]);
break;
}
- if (state == 9)
- old_state = 3;
- else
- old_state = state;
+ old_state = state;
state = -1;
out_string = out_buf;
PUT (*out_string++);
@@ -1143,10 +1115,10 @@ do_scrub_chars (size_t (*get) (char *, s
#ifdef KEEP_WHITE_AROUND_COLON
state = 9;
#else
- if (state == 9 || state == 10)
- state = 3;
- else if (state != 3)
+ if (state == 2 || state == 11)
state = 1;
+ else
+ state = 9;
#endif
PUT (ch);
break;
@@ -1271,7 +1243,7 @@ do_scrub_chars (size_t (*get) (char *, s
break;
}
-#ifdef TC_D10V
+#ifdef TC_D10V//todo drop?
/* All insns end in a char for which LEX_IS_SYMBOL_COMPONENT is true.
Trap is the only short insn that has a first operand that is
neither register nor label.
@@ -1282,7 +1254,7 @@ do_scrub_chars (size_t (*get) (char *, s
can recognize it as such. */
/* An alternative approach would be to reset the state to 1 when
we see '||', '<'- or '->', but that seems to be overkill. */
- if (state == 10)
+ if (state == 3)
PUT (' ');
#endif
/* We have a line comment character which is not at the
@@ -1296,7 +1268,7 @@ do_scrub_chars (size_t (*get) (char *, s
if (scrub_m68k_mri
&& (ch == '!' || ch == '*' || ch == '#')
&& state != 1
- && state != 10)
+ && state != 3)
goto de_fault;
/* Fall through. */
case LEX_IS_COMMENT_START:
@@ -1352,17 +1324,6 @@ do_scrub_chars (size_t (*get) (char *, s
/* Fall through. */
case LEX_IS_SYMBOL_COMPONENT:
- if (state == 10)
- {
- /* This is a symbol character following another symbol
- character, with whitespace in between. We skipped
- the whitespace earlier, so output it now. */
- UNGET (ch);
- state = 3;
- PUT (' ');
- break;
- }
-
#ifdef TC_Z80
/* "af'" is a symbol containing '\''. */
if (state == 3 && (ch == 'a' || ch == 'A'))
@@ -1388,7 +1349,16 @@ do_scrub_chars (size_t (*get) (char *, s
}
}
#endif
- if (state == 3)
+
+ /* Fall through. */
+ default:
+ de_fault:
+ /* Some relatively `normal' character. */
+ if (state == 0)
+ state = 11; /* Now seeing label definition. */
+ else if (state == 1)
+ state = 2; /* Ditto. */
+ else if (state == 3)
state = 9;
/* This is a common case. Quickly copy CH and all the
@@ -1436,52 +1406,15 @@ do_scrub_chars (size_t (*get) (char *, s
}
}
- /* Fall through. */
- default:
- de_fault:
- /* Some relatively `normal' character. */
- if (state == 0)
- {
- state = 11; /* Now seeing label definition. */
- }
- else if (state == 1)
- {
- state = 2; /* Ditto. */
- }
- else if (state == 9)
- {
- if (!IS_SYMBOL_COMPONENT (ch))
- state = 3;
- }
- else if (state == 10)
- {
- if (ch == '\\')
- {
- /* Special handling for backslash: a backslash may
- be the beginning of a formal parameter (of a
- macro) following another symbol character, with
- whitespace in between. If that is the case, we
- output a space before the parameter. Strictly
- speaking, correct handling depends upon what the
- macro parameter expands into; if the parameter
- expands into something which does not start with
- an operand character, then we don't want to keep
- the space. We don't have enough information to
- make the right choice, so here we are making the
- choice which is more likely to be correct. */
- if (to + 1 >= toend)
- {
- /* If we're near the end of the buffer, save the
- character for the next time round. Otherwise
- we'll lose our state. */
- UNGET (ch);
- goto tofull;
- }
- *to++ = ' ';
- }
+ /* As a special case, to limit the delta to previous behavior, e.g.
+ also affecting listings, go straight to state 3 when seeing a
+ comma. Commas are special: While they can be used to separate
+ macro parameters or arguments, they cannot (on their own, i.e.
+ without quoting) be arguments (or parameter default values).
+ Hence successive whitespace is not meaningful there. */
+ if (ch == ',' && state == 9)
+ state = 3;
- state = 3;
- }
PUT (ch);
break;
}
--- a/gas/config/tc-aarch64.c
+++ b/gas/config/tc-aarch64.c
@@ -641,6 +641,7 @@ const char FLT_CHARS[] = "rRsSfFdDxXeEpP
static inline bool
skip_past_char (char **str, char c)
{
+ skip_whitespace (*str);
if (**str == c)
{
(*str)++;
@@ -891,6 +892,7 @@ parse_reg (char **ccp)
start++;
#endif
+ skip_whitespace (start);
p = start;
if (!ISALPHA (*p) || !is_name_beginner (*p))
return NULL;
@@ -1196,13 +1198,17 @@ parse_typed_reg (char **ccp, aarch64_reg
struct vector_type_el *typeinfo, unsigned int flags)
{
char *str = *ccp;
- bool is_alpha = ISALPHA (*str);
- const reg_entry *reg = parse_reg (&str);
+ bool is_alpha;
+ const reg_entry *reg;
struct vector_type_el atype;
struct vector_type_el parsetype;
bool is_typed_vecreg = false;
unsigned int err_flags = (flags & PTR_IN_REGLIST) ? SEF_IN_REGLIST : 0;
+ skip_whitespace (str);
+ is_alpha = ISALPHA (*str);
+ reg = parse_reg (&str);
+
atype.defined = 0;
atype.type = NT_invtype;
atype.width = -1;
@@ -1420,10 +1426,7 @@ parse_vector_reg_list (char **ccp, aarch
do
{
if (in_range)
- {
- str++; /* skip over '-' */
- val_range = val;
- }
+ val_range = val;
const reg_entry *reg;
if (has_qualifier)
@@ -1485,7 +1488,8 @@ parse_vector_reg_list (char **ccp, aarch
in_range = 0;
ptr_flags |= PTR_GOOD_MATCH;
}
- while (skip_past_comma (&str) || (in_range = 1, *str == '-'));
+ while (skip_past_comma (&str)
+ || (in_range = 1, skip_past_char (&str, '-')));
skip_whitespace (str);
if (*str != '}')
@@ -8265,6 +8269,7 @@ parse_operands (char *str, const aarch64
}
/* Check if we have parsed all the operands. */
+ skip_whitespace (str);
if (*str != '\0' && ! error_p ())
{
/* Set I to the index of the last present operand; this is
--- a/gas/config/tc-arm.c
+++ b/gas/config/tc-arm.c
@@ -1148,6 +1148,8 @@ my_get_expression (expressionS * ep, cha
prefix_mode = (prefix_mode == GE_OPT_PREFIX_BIG) ? prefix_mode
: GE_OPT_PREFIX;
+ skip_whitespace (*str);
+
switch (prefix_mode)
{
case GE_NO_PREFIX: break;
--- a/gas/config/tc-i386.c
+++ b/gas/config/tc-i386.c
@@ -15306,6 +15306,8 @@ RC_SAE_immediate (const char *imm_start)
as_bad (_("Missing '}': '%s'"), imm_start);
return 0;
}
+ if (is_space_char (*pstr))
+ ++pstr;
/* RC/SAE immediate string should contain nothing more. */;
if (*pstr != 0)
{
More information about the Binutils
mailing list