diff options
| -rw-r--r-- | subx/033non_code_segment.cc | 30 | ||||
| -rw-r--r-- | subx/034discourage_raw_hex.cc | 36 | ||||
| -rw-r--r-- | subx/035labels.cc | 4 | ||||
| -rw-r--r-- | subx/036recommend_labels.cc | 81 | ||||
| -rw-r--r-- | subx/037label_types.cc | 45 | ||||
| -rw-r--r-- | subx/038check_local_jumps.cc | 60 | ||||
| -rw-r--r-- | subx/subx.vim | 17 |
7 files changed, 4 insertions, 269 deletions
diff --git a/subx/033non_code_segment.cc b/subx/033non_code_segment.cc deleted file mode 100644 index c0549023..00000000 --- a/subx/033non_code_segment.cc +++ /dev/null @@ -1,30 +0,0 @@ -//: Raise an error when operand metadata is used in non-code segments. - -:(scenario operand_metadata_outside_code_segment) -% Hide_errors = true; -== 0x1 # code segment -cd 0x80/imm8 -== 0x1000 # data segment -cd 12/imm8 -+error: 12/imm8: metadata imm8 is only allowed in the (first) code segment - -:(before "Pack Operands(segment code)") -ensure_operands_only_in_code_segments(p); -if (trace_contains_errors()) return; -:(code) -void ensure_operands_only_in_code_segments(const program& p) { - trace(99, "transform") << "-- ensure operands only in code segments" << end(); - if (p.segments.empty()) return; - for (int i = /*skip code segment*/1; i < SIZE(p.segments); ++i) { - const segment& seg = p.segments.at(i); - for (int j = 0; j < SIZE(seg.lines); ++j) { - const line& l = seg.lines.at(j); - for (int k = 0; k < SIZE(l.words); ++k) { - const word& w = l.words.at(k); - for (map<string, uint32_t>::iterator p = Operand_bound.begin(); p != Operand_bound.end(); ++p) - if (has_metadata(w, p->first)) - raise << w.original << ": metadata " << p->first << " is only allowed in the (first) code segment\n" << end(); - } - } - } -} diff --git a/subx/034discourage_raw_hex.cc b/subx/034discourage_raw_hex.cc deleted file mode 100644 index 2b2e2675..00000000 --- a/subx/034discourage_raw_hex.cc +++ /dev/null @@ -1,36 +0,0 @@ -//: Now that we have operand metadata, start warning on instructions that -//: don't use it. -//: -//: While SubX will let you write raw machine code, don't do that unless you -//: have a very good reason. - -:(before "Pack Operands(segment code)") -warn_on_raw_hex(code); -if (trace_contains_errors()) return; -:(code) -void warn_on_raw_hex(const segment& code) { - trace(99, "transform") << "-- warn on raw hex instructions" << end(); - for (int i = 0; i < SIZE(code.lines); ++i) { - const line& inst = code.lines.at(i); - if (all_hex_bytes(inst) && has_operands(inst)) { - warn << "'" << to_string(inst) << "': using raw hex is not recommended\n" << end(); - break; - } - } -} - -:(scenarios transform) -:(scenario warn_on_hex_bytes_without_operands) -== 0x1 -bb 2a 00 00 00 # copy 0x2a (42) to EBX -+warn: 'bb 2a 00 00 00': using raw hex is not recommended - -:(scenario warn_on_non_operand_metadata) -== 0x1 -bb 2a 00/foo 00/bar 00 # copy 0x2a (42) to EBX -+warn: 'bb 2a 00/foo 00/bar 00': using raw hex is not recommended - -:(scenario no_warn_on_instructions_without_operands) -== 0x1 -55 # push EBP --warn: '55': using raw hex is not recommended diff --git a/subx/035labels.cc b/subx/035labels.cc index 236c8fa2..0d366d2e 100644 --- a/subx/035labels.cc +++ b/subx/035labels.cc @@ -1,5 +1,9 @@ //: Labels are defined by ending names with a ':'. This layer will compute //: addresses for labels, and compute the offset for instructions using them. +//: +//: We won't check this, but our convention will be that jump targets will +//: start with a '$', while functions will not. Function names will never be +//: jumped to, and jump targets will never be called. //: We're introducing non-number names for the first time, so it's worth //: laying down some ground rules all transforms will follow, so things don't diff --git a/subx/036recommend_labels.cc b/subx/036recommend_labels.cc deleted file mode 100644 index 3d603842..00000000 --- a/subx/036recommend_labels.cc +++ /dev/null @@ -1,81 +0,0 @@ -//: Now that we have labels, using non-label offsets should be unnecessary. -//: While SubX will allow programmers to write raw machine code, that isn't -//: *recommended* once we have more ergonomic alternatives. - -:(scenario warn_on_jump_offset) -== 0x1 -7e/jump-if 1/disp8 -+warn: '7e/jump-if 1/disp8': using raw offsets for jumps is not recommended; use labels instead - -:(scenarios transform) -:(scenario warn_on_call_offset) -== 0x1 -e8/call 1/disp32 -+warn: 'e8/call 1/disp32': using raw offsets for calls is not recommended; use labels instead -:(scenarios run) - -:(before "Rewrite Labels(segment code)") -recommend_labels(code); -if (trace_contains_errors()) return; -:(code) -void recommend_labels(const segment& code) { - trace(99, "transform") << "-- check for numeric labels" << end(); - for (int i = 0; i < SIZE(code.lines); ++i) - recommend_labels(code.lines.at(i)); -} - -void recommend_labels(const line& inst) { - int idx = first_operand(inst); - if (idx >= SIZE(inst.words)) return; - if (!is_number(inst.words.at(idx).data)) return; - if (is_jump(inst)) - warn << "'" << inst.original << "': using raw offsets for jumps is not recommended; use labels instead\n" << end(); - else if (is_call(inst)) - warn << "'" << inst.original << "': using raw offsets for calls is not recommended; use labels instead\n" << end(); -} - -bool is_jump(const line& inst) { - string op1 = preprocess_op(inst.words.at(0)).data; - if (op1 == "0f") { - string op2 = preprocess_op(inst.words.at(1)).data; - return Jump_opcodes_0f.find(op1) != Jump_opcodes_0f.end(); - } - if (op1 == "ff") return subop(inst) == /*subop for opcode ff*/4; - return Jump_opcodes.find(op1) != Jump_opcodes.end(); -} - -bool is_call(const line& inst) { - string op1 = preprocess_op(inst.words.at(0)).data; - if (op1 == "e8") return true; - if (op1 == "ff") return subop(inst) == /*subop for opcode ff*/2; - return false; // no multi-byte call opcodes -} - -int subop(const line& inst) { - int idx = first_operand(inst); - assert(idx < SIZE(inst.words)); - return (parse_int(inst.words.at(idx).data)>>3) & 0x7; -} - -:(before "End Globals") -set<string> Jump_opcodes; -set<string> Jump_opcodes_0f; -:(before "End One-time Setup") -init_jump_opcodes(); -:(code) -void init_jump_opcodes() { - Jump_opcodes.insert("74"); - Jump_opcodes.insert("75"); - Jump_opcodes.insert("7c"); - Jump_opcodes.insert("7d"); - Jump_opcodes.insert("7e"); - Jump_opcodes.insert("7f"); - Jump_opcodes_0f.insert("84"); - Jump_opcodes_0f.insert("85"); - Jump_opcodes_0f.insert("8c"); - Jump_opcodes_0f.insert("8d"); - Jump_opcodes_0f.insert("8e"); - Jump_opcodes_0f.insert("8f"); - Jump_opcodes.insert("e9"); - Jump_opcodes.insert("eb"); -} diff --git a/subx/037label_types.cc b/subx/037label_types.cc deleted file mode 100644 index b80db732..00000000 --- a/subx/037label_types.cc +++ /dev/null @@ -1,45 +0,0 @@ -//: Distinguish between labels marking the start of a function, and labels -//: inside functions. -//: -//: - Labels within functions start with a '$', and are only permitted in -//: 'jump' instructions. -//: -//: - Labels marking the start of functions lack the '$' sigil, and are only -//: permitted in 'call' instructions. - -:(before "Rewrite Labels(segment code)") -check_label_types(code); -if (trace_contains_errors()) return; -:(code) -void check_label_types(const segment& code) { - trace(99, "transform") << "-- check label types" << end(); - for (int i = 0; i < SIZE(code.lines); ++i) - check_label_types(code.lines.at(i)); -} - -void check_label_types(const line& inst) { - int idx = first_operand(inst); - if (idx >= SIZE(inst.words)) return; - const word& target = inst.words.at(idx); - if (is_number(target.data)) return; // handled elsewhere - if (is_jump(inst) && target.data.at(0) != '$') - raise << "'" << inst.original << "': jumps should always be to internal labels starting with '$'\n" << end(); - if (is_call(inst) && target.data.at(0) == '$') - raise << "'" << inst.original << "': calls should always be to function labels (not starting with '$')\n" << end(); -} - -:(scenario catch_jump_to_function) -% Hide_errors = true; -== 0x1 -main: -7e/jump-if foo/disp8 -foo: -+error: '7e/jump-if foo/disp8': jumps should always be to internal labels starting with '$' - -:(scenario catch_call_to_internal_label) -% Hide_errors = true; -== 0x1 -main: -e8/call $foo/disp32 - $foo: # indent to avoid looking like a trace_count command for this scenario -+error: 'e8/call $foo/disp32': calls should always be to function labels (not starting with '$') diff --git a/subx/038check_local_jumps.cc b/subx/038check_local_jumps.cc deleted file mode 100644 index 41f8e471..00000000 --- a/subx/038check_local_jumps.cc +++ /dev/null @@ -1,60 +0,0 @@ -//: Make sure that we never jump from one function to within another. -//: -//: (The check for label types already ensures we can't jump to the start of -//: another function.) - -:(scenario jump_to_different_function) -% Hide_errors = true; -== 0x1 -fn1: - 7e/jump-if $target/disp8 -fn2: - $target: -+error: '7e/jump-if $target/disp8' in function 'fn1': jump to within another function 'fn2' is a *really* bad idea - -:(before "Rewrite Labels(segment code)") -check_local_jumps(code); -if (trace_contains_errors()) return; -:(code) -void check_local_jumps(const segment& code) { - map</*jump target*/string, /*containing call target*/string> function; - compute_function_target(code, function); - if (trace_contains_errors()) return; - string current_function; - for (int i = 0; i < SIZE(code.lines); ++i) { - const line& inst = code.lines.at(i); - if (SIZE(inst.words) == 1 && is_label(inst.words.at(0))) { - // label definition - if (inst.words.at(0).data.at(0) != '$') - current_function = drop_last(inst.words.at(0).data); - } - else if (is_jump(inst)) { - const word& target = inst.words.at(first_operand(inst)); - if (!contains_key(function, target.data)) continue; // error/warning handled elsewhere - if (get(function, target.data) == current_function) continue; - raise << "'" << to_string(inst) << "' in function '" << current_function << "': jump to within another function '" << get(function, target.data) << "' is a *really* bad idea\n" << end(); - return; - } - } -} - -void compute_function_target(const segment& code, map<string, string>& out) { - string current_function; - for (int i = 0; i < SIZE(code.lines); ++i) { - const line& inst = code.lines.at(i); - if (SIZE(inst.words) != 1) continue; - const word& curr = inst.words.at(0); - if (!is_label(curr)) continue; - const string& label = drop_last(curr.data); - if (label.at(0) != '$') { - current_function = label; - continue; - } - if (contains_key(out, label)) { - raise << "duplicate label '" << label << "'\n" << end(); - return; - } - // current_function can be empty! if so that would be 'main'. - put(out, label, current_function); - } -} diff --git a/subx/subx.vim b/subx/subx.vim index e6825c72..454c9454 100644 --- a/subx/subx.vim +++ b/subx/subx.vim @@ -25,21 +25,4 @@ let b:cmt_head = "#? " " comment token syntax match subxDelimiter / \. / | highlight link subxDelimiter Delimiter -"" highlight low-level idioms in red as I provide more high-level replacements - -" Once we have labels, highlight raw displacement -highlight Warn ctermbg=brown ctermfg=black -call matchadd("Warn", '\c^\s*e8.*\<\(0x\)\?[0-9a-f]\+/disp32') " call -call matchadd("Warn", '\c^\s*e9.*\<\(0x\)\?[0-9a-f]\+/disp8') " unconditional jump disp8 -call matchadd("Warn", '\c^\s*7[45cdef].*\<\(0x\)\?[0-9a-f]\+/disp8') " conditional jump disp8 -call matchadd("Warn", '\c^\s*eb.*\<\(0x\)\?[0-9a-f]\+/disp16') " unconditional jump disp16 -call matchadd("Warn", '\c^\s*0f[^\s]*\s*8[45cdef].*\<\(0x\)\?[0-9a-f]\+/disp16') " conditional jump disp16 - -" Mismatch in label type -call matchadd("Error", '\c^\s*e8.*\$') " call -call matchadd("Error", '\c^\s*e9\(/[^ ]*\)\?\s*\(\.\s*\)\+[^\$\. ]\([ \$0-9a-fA-F-]\+\>\)\@!') " unconditional jump disp8 -call matchadd("Error", '\c^\s*7[45cdef]\(/[^ ]*\)\?\s*\(\.\s*\)\+[^\$\. ]\([ \$0-9a-fA-F-]\+\>\)\@!') " conditional jump disp8 -call matchadd("Error", '\c^\s*eb\(/[^ ]*\)\?\s*\(\.\s*\)\+[^\$\. ]\([ \$0-9a-fA-F-]\+\>\)\@!') " unconditional jump disp16 -call matchadd("Error", '\c^\s*0f[^\s]*\s*8[45cdef]\(/[^ ]*\)\?\s*\(\.\s*\)\+[^\$\. ]\([ \$0-9a-fA-F-]\+\>\)\@!') " conditional jump disp16 - let &cpo = s:save_cpo |