Commit graph

1370 commits

Author SHA1 Message Date
Bill Currie
33810e9c91 Write out object files in the new format. 2011-02-23 15:05:33 +09:00
Bill Currie
1f7c39b918 Chain @va_list's type.
Oops :P
2011-02-23 14:49:00 +09:00
Bill Currie
a1ea492889 Remove some redundant fields from qfo objects. 2011-02-23 10:40:35 +09:00
Bill Currie
c585a17db9 Give type data a null pointer.
Pointers to the void type descriptor looked just like null pointers and
that was terribly confusing. Avoid that confusion by pre-allocating a
small amount of data in order to reserve space for a null descriptor.
2011-02-23 08:08:26 +09:00
Bill Currie
e68e6d0858 Create defs for entity fields in the entity space. 2011-02-22 15:43:15 +09:00
Bill Currie
e1158a83f0 Begin work on rewriting obj_file.c for the new format. 2011-02-22 15:42:49 +09:00
Bill Currie
64cf07351f Encode types to string and qfo representation on creation. 2011-02-22 13:36:16 +09:00
Bill Currie
b50b392746 Fix field def allocation for function fields.
Assignment of nil to a field function is permitted, but trying to use one
as a builtin or as a normal function is treated as an error.
.void (float y) func;       OK
.void (float y) bi = #0;    error
.void (float y) ni = nil;   OK
.void (float y) co = { };   error
2011-02-22 12:07:31 +09:00
Bill Currie
7d1f2c4ef3 Fix field def allocation for non-function fields. 2011-02-22 10:43:35 +09:00
Bill Currie
b7c6c4af5d Better variable naming in init_elements(). 2011-02-22 09:22:04 +09:00
Bill Currie
497db3ac81 Fix the return symbol extraction when appending a return statement.
The code is still icky, though :P
2011-02-22 09:20:52 +09:00
Bill Currie
4fd6bdae6f Allow nil to be assigned to field variables.
Note, this is the field variable itself, not the field in the entity.
eg:
	.integer x = nil;
2011-02-22 09:19:01 +09:00
Bill Currie
87b240f469 Use special treatment for .float ()foo; functions.
The parser wants to treat .float () foo; as a function returning a float
field, but qcc treats it as a field holding a function variable.
Fortuantely, field types are always "simple" (ie, at worst, just more
field type wrappers around the non-field type), so all that's needed to
obtain qcc grammar is to reach into the field type layers and do the
function type calculation based on the non-field type found there.
2011-02-22 08:54:01 +09:00
Bill Currie
4c11304c8d Documentation tweaks. 2011-02-21 08:41:03 +09:00
Bill Currie
bffc1dd239 Clean up objective-qc type encoding.
Structures etc now never encode the fields as well (I might revert that
at some time). Type parsing has been removed: it has proven unnecessary
and overly complicated, and the new qfo encoding should be more useful.
2011-02-18 20:51:04 +09:00
Bill Currie
41523852cd Cleaner def and string emittion for types.
Use local defines for emitting defs and strings as everything goes to the
same locations.
2011-02-18 12:44:11 +09:00
Bill Currie
7251bc91fa Enhanced type encoding.
This creates a compact representation of types as seen by qfcc. This is
intended mainly for debuggers.
2011-02-18 09:06:05 +09:00
Bill Currie
0f0ed3dae2 Allow null types in defs again.
However, only at the lowest level. This is needed for my type encoding.
2011-02-18 09:04:33 +09:00
Bill Currie
bceb98e5a4 Create structures for new qfo format.
This breaks a lot of existing code, so that code has been commented out
for now.
2011-02-17 15:02:09 +09:00
Bill Currie
9c9a71f1af Allow defs to specify use of offset relocations.
Access to struct fields in near data can be done using only one operand,
but offset relocs need to be used. However, as not all defs want offset
relocs, a flag has been added to the def struct.
2011-02-15 22:55:15 +09:00
Bill Currie
d04865884f Set the method's function pointer. 2011-02-15 22:27:13 +09:00
Bill Currie
3721b2edca Method lists are made of methods, not integers :P 2011-02-15 22:26:44 +09:00
Bill Currie
c068138d91 More addresses in hex. 2011-02-15 22:26:09 +09:00
Bill Currie
814e36d3c8 Correct an error message and tidy some code. 2011-02-15 22:25:40 +09:00
Bill Currie
4336fc2c73 Move method function defs into the far data space.
As they are never referenced directly by instructions, there's no need for
them to be in the near data space, taking precious def locations.
2011-02-15 15:33:18 +09:00
Bill Currie
5252d72856 Correct the type for _OBJ_SELECTOR_TABLE. 2011-02-15 15:33:12 +09:00
Bill Currie
b3da6ef3d4 Allow zero size arrays to match sized arrays of the same type.
This allows externally declared arrays to not specify a size.
2011-02-15 12:08:03 +09:00
Bill Currie
b2d4aee61e Reimplement selector expressions to work with far data.
They currently execute a lea instruction every method call, but the
optimizer should take care of that once its written.
2011-02-15 10:48:05 +09:00
Bill Currie
6bd63f811f Preserve any relocs when converting a def from extern to otherwise. 2011-02-15 10:47:16 +09:00
Bill Currie
4e86b28b59 Use the def type rather than expression type to select the init method.
Local defs may be initialized with non-constant expressions, which
includes function calls. Both function calls and aggregate initializers
use block expressions.
2011-02-15 09:38:44 +09:00
Bill Currie
19f6faf0ad Fix aggregate initializers.
Initializing arrays and structs seems to be working.
2011-02-15 09:30:37 +09:00
Bill Currie
d162838299 Resurrect constant_expr(). 2011-02-15 09:28:27 +09:00
Bill Currie
bc882ddc3c Fix the parser side of aggregate initializers. 2011-02-14 23:31:37 +09:00
Bill Currie
09af401f0c Reset current_func at the end of each function. 2011-02-14 23:10:46 +09:00
Bill Currie
5fa349927f Preserve the ivars inheritance chain across build_struct().
build_struct() unlinks the structure symbol table from its parent, but
that breaks the ivar inheritance chain.
2011-02-14 23:10:46 +09:00
Bill Currie
5707888fcf Fix some error-recovery segfaults. 2011-02-14 23:10:46 +09:00
Bill Currie
f1fccb19dd Allow same-type re-declarations of uninitialized variables.
This is especially important for declaring variables that were previously
declared external.
2011-02-14 23:10:46 +09:00
Bill Currie
23b01be214 Handle invalid initializers gracefully. 2011-02-14 23:10:46 +09:00
Bill Currie
4c63a0bad6 Generate a warning when creating a var with no type.
This makes the internal error in new_def() non-functional, but the warning
will catch the bug and there's no harm in leaving the test in new_def().
2011-02-14 23:10:46 +09:00
Bill Currie
ee866ae3a6 Enable the small struct optimization.
For now, anyway. It would probably be much better to handle this sort of
thing in the optimizer stage.
2011-02-14 23:10:45 +09:00
Bill Currie
bcaf3687c5 Fix loading structs into params via an offset pointer.
For certain values of "fix" :/. The code is ugly, but it does the right
thing: calculate the effect address and use the resulting pointer in a
move instruction.
2011-02-14 23:10:45 +09:00
Bill Currie
352be1f8bd Handle null string pointers. 2011-02-14 23:10:45 +09:00
Bill Currie
e10d561459 Treat small structs as integers (disabled).
This is more to stash away a potential optimization. I need to fix the
real problem first.
2011-02-14 23:10:45 +09:00
Bill Currie
6b29568fa9 Set the source file and line when copying an expression.
This fixes the crazy line numbers in the debug info.
2011-02-14 23:10:45 +09:00
Bill Currie
fc18ebd923 Add the source line number to most node labels. 2011-02-14 23:10:45 +09:00
Bill Currie
3e5989463f Update the def_tail when removing the last def from the space. 2011-02-14 23:10:45 +09:00
Bill Currie
11485556e5 Quote expression names to avoid problems with null pointers. 2011-02-14 23:10:45 +09:00
Bill Currie
c139099294 Put another bandaid on the type creation system.
It seems that building types is very fragile: all too easy to fix things
in one place and break something somewhere else. I guess the type system
will need a comlete rewrite at some stage.
2011-02-14 23:10:45 +09:00
Bill Currie
e0d61705b4 And fix build_struct for built-in struct types.
The previous fix broke builtin structs (eg, @param). Calling find_type()
for built-in structs has proven to be a bad idea :)
2011-02-13 21:07:22 +09:00
Bill Currie
a19b7ec8f9 Make struct decls actually work.
find_type() does type comparisons based on tag name for struct/union/enum
but when it returns the previous tag (eg, from "struct foo;") the returned
type has a null symtab pointer, so setting the symtab pointer in the type
before calling find_type() only throws away the symtab. Thus, since
find_type() doesn't check the symtab pointer when comparing the types, set
the symtab pointer after calling find_type().
2011-02-13 20:56:12 +09:00
Bill Currie
acced90101 Create the symbol for the struct as early as possible. 2011-02-13 20:55:33 +09:00
Bill Currie
b545f72f93 Report errors for incomplete types and missing fields. 2011-02-13 20:23:08 +09:00
Bill Currie
c336aaeb80 Handle incomplete types gracefully. 2011-02-13 19:26:20 +09:00
Bill Currie
976ff8a6bb Allow cast expressions to be lvalues.
These really need to be alias expressions rather than cast expressions
so the two can be separated as cast expressions should not be lvalues.
2011-02-13 17:54:15 +09:00
Bill Currie
fa742ceed6 Handle @self and @this correctly. 2011-02-13 17:52:19 +09:00
Bill Currie
547f361bd3 Allow (void) qc-style functions. 2011-02-13 17:37:11 +09:00
Bill Currie
d75c35ba30 Allow qc style function params.
Due to ambiguities in the grammar, qc-style function params and c-style
function params had to be completely separated. This means that qc-style
functions can not use pointers and must use qc-style function declarations
for parameters, and c-style functions must use c-style param declarations.
While this rule is tedious for converting the Ruamoko library, it does
actually make for a more consistent language.
2011-02-13 17:07:14 +09:00
Bill Currie
8a314e9e87 Class pointers need to be directly addressable. 2011-02-13 16:05:50 +09:00
Bill Currie
0b5565396b Add a class type to symbols and rework the CLASS_NAME handling.
This fixes the problem with [Array alloc] producing a warning about Class
not repsonding to -alloc.
2011-02-13 16:05:09 +09:00
Bill Currie
3f690eacb1 Fix another double insertion of a symbol. 2011-02-13 15:28:49 +09:00
Bill Currie
127a1c8772 Fix the creation of enum values. 2011-02-13 15:19:55 +09:00
Bill Currie
f24de89a2a Handle type-modified constants.
The fix for pointers broke constants (particularly, enums). Oops.
2011-02-13 14:13:32 +09:00
Bill Currie
fa6b8d773c Correctly emit class and category symtab references. 2011-02-13 13:21:00 +09:00
Bill Currie
c4f364fd37 Emit pointers properly.
Rename ReuseConstant to emit_value and use emit_value to emit any non-short
constant. This fixes the null pointer in the call to __obj_exec_class.
2011-02-13 10:09:42 +09:00
Bill Currie
1bcf434b20 Move _OBJ_MODULE to the far data space. 2011-02-12 23:00:04 +09:00
Bill Currie
d42a379924 Dump the contents of a pointer def. 2011-02-12 22:59:29 +09:00
Bill Currie
eede83dea1 Avoid crashing when a null class pointer is found. 2011-02-12 22:58:50 +09:00
Bill Currie
3505de6868 Point the near and far data spaced into the merged data space.
This lets the relocations work on the final data without having to adjust
their locations.
2011-02-12 22:35:44 +09:00
Bill Currie
181596f5bf Make relocation records store the space of the relocation.
The space is meaningless for op_* relocations as they are always in the
code space, but def_* relocations need to know which space holds the
location to be adjusted.
2011-02-12 22:34:38 +09:00
Bill Currie
8c01497936 Give __obj_exec_class a sterner beating.
When linking a def into a defspace, ensure it can be removed. This fixes
the missing __obj_exec_class def in qwaq.
2011-02-11 22:29:57 +09:00
Bill Currie
75d90d2f00 Propogate the specifiers through external_decl_list. 2011-02-11 22:29:57 +09:00
Bill Currie
0efc759f2d Catch attempts to create a def with a null type. 2011-02-11 22:29:57 +09:00
Bill Currie
5f3c1872b9 Put class and category reference/name symbols in the symbol table.
This fixes the undefined defs in qwaq.
2011-02-11 22:29:57 +09:00
Bill Currie
179c1f7058 Support move statements (structure copy). 2011-02-11 22:29:57 +09:00
Bill Currie
e19f7e3521 Sort out the parameters for QC style functions :) 2011-02-11 22:29:57 +09:00
Bill Currie
370f0b97ed Produce cleaner output for a = b op c. 2011-02-11 22:29:57 +09:00
Bill Currie
d93d8d7d46 Remove the dirty hack used for accessing params and the return value.
Instead of using the equivalent of *(float*)&.return, now use the
equivalent of (float).return. No conversion is done in the "cast".

NOTE: this sort of cast should be separated from normal casts.
2011-02-11 22:29:38 +09:00
Bill Currie
f5412c2b20 Remove a fixme. 2011-02-11 22:29:38 +09:00
Bill Currie
c185fbb782 Set the type of the super structure to the correct type. 2011-02-11 22:29:38 +09:00
Bill Currie
7d2a95ee40 Mark a few problem spots. 2011-02-11 22:29:37 +09:00
Bill Currie
a1d4d4bae1 Handle block expressions in is_indirect(). 2011-02-11 22:29:30 +09:00
Bill Currie
b01f960db9 Disable the subgraphs for now.
They tend to make large trees difficult to read. I might re-enable them
when I've gotten a better handle on dot.
2011-02-10 15:11:08 +09:00
Bill Currie
c3a991e555 Support nil in initializers. 2011-02-10 14:51:52 +09:00
Bill Currie
adeef4a4eb Copy named expressions.
Since expressions can be modified, making DAGs this early causes
corruption of the expression tree.
2011-02-10 14:43:49 +09:00
Bill Currie
6f625e426f Use the correct type when deferencing an offset pointer. 2011-02-10 14:26:12 +09:00
Bill Currie
1099ca086c Dump the offending expression and statement for a bad instruction. 2011-02-10 14:25:09 +09:00
Bill Currie
4a41d97afc Angle brackets and html :P 2011-02-10 14:24:31 +09:00
Bill Currie
01a7af17e5 Beat __obj_exec_class into submission. 2011-02-10 00:15:19 +09:00
Bill Currie
006c16807d Add the statement for based loads.
Oops :P
2011-02-09 22:40:16 +09:00
Bill Currie
a43946abf3 Give short values special treatment.
As a short value fits directly into a statement, it needs a fake def that
holds its value in the def's offset in order for the value to be written to
the statement.
2011-02-09 22:13:43 +09:00
Bill Currie
c7adde8709 Fix some case&paste errors. 2011-02-09 22:04:54 +09:00
Bill Currie
1b8edc07c9 Adjust the offsets of ivars in derived classes.
As ivar structs are not independent, their starting offset is not
necessarily zero (almost never will be for derived classes).
2011-02-09 22:02:52 +09:00
Bill Currie
d66a441867 Write out the far data space as well as the near.
This causes the Objective-QC structs to be emitted now that they havve been
moved to the far data space. Relocs are broken, though.
2011-02-09 21:45:29 +09:00
Bill Currie
244d969ae3 Ensure any tail bytes in the strings data are zeroed. 2011-02-09 18:32:37 +09:00
Bill Currie
2d5d8eda5a Save the caller's address when creating new relocations. 2011-02-09 17:16:49 +09:00
Bill Currie
29c1f7ac03 Use the correct type object in emit_struct(). 2011-02-09 17:15:29 +09:00
Bill Currie
59a683d39e Protect the name passed into emit_struct().
The name passed into emit_struct is sometimes created via va(), but the
struct creation code also uses va. This makes a mess of names.
2011-02-09 17:12:47 +09:00
Bill Currie
3e1b034aaf Use EMIT_DEF correctly. 2011-02-09 15:58:32 +09:00
Bill Currie
e75295b00c Make normal C style functions generate code. 2011-02-09 10:25:23 +09:00
Bill Currie
68cdc904f5 Give temporary variables unique names. 2011-02-09 10:04:55 +09:00