1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
|
Write test for empty_char_constant
defined cannot be used as a macro name
<strike>Add "defined" and only accept it in appropriate circumstances</strike>
Update that simple tokenizer compulsory test so things will compile
Handle cases like escaped question marks and pound symbols that I don't understand yet.
(done) Fix #include <stdio.h> to read include directive correctly
txt/orig state of affairs:
The problem is that there are two ways to interpret line,col:
With respect to txt
With respect to orig
This isn't a problem when txt and orig point to the same character, as in:
int in\
dex
int \
index /*Here, the backslash break should be gobbled up by the space identifier*/
line,col has no ambiguity as to where it should point. However, when they point to different characters (i.e. at the beginning of a line):
\
int index
line,col could either point to orig or to the first real character. Thus, we will do the latter.
Moreover, will a newline followed by backslash breaks generate a token that gobbles up said breaks? I believe it will, but no need to call this mandatory.
Thus, on a lookup with a txt pointer, the line/col/orig should match the real character and not preceding backslash breaks.
I've been assuming that every token starts with its first character, neglecting the case where a line starts with backslash breaks. The question is, given the txt pointer to the first character, where should the derived orig land?
Currently, the orig lands after the beginning backslash breaks, when instead it should probably land before them.
Here's what the tokenizer's text anchoring needs:
Broken/unbroken text pointer -> line/col
Unbroken contents per token to identify identifier text
Original contents per token to rebuild the document
Ability to change "original contents" so the document will be saved with modifications
Ability to insert new tokens
Solution:
New tokens will typically have identical txt and orig, yea even the same pointer.
txt/txt_size for unbroken contents, orig/orig_size for original
modify orig to change the document
txt identifies identifier text
Line lookup tables are used to resolve txt/orig pointers; other pointers can't be resolved in the same fashion and may require traversing backward through the list.
What this means:
Token txt/txt_size, orig/orig_size, orig_lines, txt_lines, and tok_point_lookup are all still correct.
Token line,col will be removed
Other improvements to do:
Sanity check the point lookups like crazy
Remove the array() structures in token_list, as these are supposed to be read-only
Make sure tok_point_lookup returns correct values for every single pointer possible, particularly those in orig that are on backslash-breaks
Convert the tok_message_queue into an array of messages bound to tokens.
Ask Rusty about the trailing newline in this case:
/* Blah
*
* blah
*/
Here, rather than the trailing space being blank, it is "blank" from the comment perspective.
May require deeper analysis.
Todos from ccan_tokenizer.h
/*
Assumption: Every token fits in one and exactly one line
Counterexamples:
Backslash-broken lines
Multiline comments
Checks to implement in the tokenizer:
is the $ character used in an identifier (some configurations of GCC allow this)
are there potentially ambiguous sequences used in a string literal (e.g. "\0000")
Are there stray characters? (e.g. '\0', '@', '\b')
Are there trailing spaces at the end of lines (unless said spaces consume the entire line)?
Are there trailing spaces after a backslash-broken line?
Fixes todo:
backslash-newline sequence should register as an empty character, and the tokenizer's line value should be incremented accordingly.
*/
Lex angle bracket strings in #include
Check the rules in the documentation
Examine the message queue as part of testing the tokenizer:
Make sure there are no bug messages
Make sure files compile with no warnings
For the tokenizer sanity check, make sure integers and floats have valid suffixes respectively
(e.g. no TOK_F for an integer, no TOK_ULL for a floating)
Update the scan_number sanity checks
(done) Move scan_number et al. to a separate C file
Test:
Overflow and underflow floats
0x.p0
(done) 0755f //octal 0755 with invalid suffix
(done) 0755e1 //floating 7550
Figure out how keywords will be handled.
Preprocessor directives are <strike>case-insensitive</strike> actually case-sensitive (except __VA_ARGS__)
All C keywords are case sensitive
__VA_ARGS__ should be read as an identifier unless it's in the expansion of a macro. Otherwise, GCC generates a warning.
We are in the expansion of a macro after <startline> <space> # <space>
Don't forget about __attribute__
Except for __VA_ARGS__, all preprocessor keywords are proceeded by <startline> <space> # <space>
Solution:
All the words themselves will go into one opkw dictionary, and for both type and opkw, no distinction will be made between preprocessor and normal keywords.
Instead, int type will become short type; unsigned short cpp:1;
Merge
Commit ccan_tokenizer to the ccan repo
Introduce ccan_tokenizer to ccanlint
Write testcases for scanning all available operators
Support integer and floating point suffices (e.g. 500UL, 0.5f)
Examine the message queue after tokenizing
Make sure single-character operators have an opkw < 128
Make sure c_dictionary has no duplicate entries
Write verifiers for other types than TOK_WHITE
What's been done:
Operator table has been organized
Merged Rusty's changes
Fixed if -> while in finalize
Fixed a couple mistakes in run-simple-token.c testcases themselves
Expected orig/orig_size sizes weren't right
Made token_list_sanity_check a public function and used it throughout run-simple-token.c
Tests succeed and pass valgrind
Lines/columns of every token are recorded
(done) Fix "0\nstatic"
(done) Write tests to make sure backslash-broken lines have correct token locations.
(done) Correctly handle backslash-broken lines
One plan: Separate the scanning code from the reading code. Scanning sends valid ranges to reading, and reading fills valid tokens for the tokenizer/scanner to properly add
Another plan: Un-break backslash-broken lines into another copy of the input. Create an array of the positions of each real line break so
Annotate message queue messages with current token
Conversion to make:
From:
Position in unbroken text
To:
Real line number
Real offset from start of line
Thus, we want an array of real line start locations wrt the unbroken text
Here is a bro\
ken line. Here is a
real line.
<LINE>Here is a bro<LINE>ken line. Here is a
<LINE>real line.
If we know the position of the token text wrt the unbroken text, we can look up the real line number and offset using only the array of real line start positions within the unbroken text.
Because all we need is the orig and orig_size with respect to the unbroken text to orient
|
