xref: /netbsd-src/usr.bin/xlint/lint1/README.md (revision ae082add65442546470c0ba499a860ee89eed305)
1[//]: # ($NetBSD: README.md,v 1.9 2022/07/08 20:27:36 rillig Exp $)
2
3# Introduction
4
5Lint1 analyzes a single translation unit of C code.
6
7* It reads the output of the C preprocessor, retaining the comments.
8* The lexer in `scan.l` and `lex.c` splits the input into tokens.
9* The parser in `cgram.y` creates types and expressions from the tokens.
10* It checks declarations in `decl.c`.
11* It checks initializations in `init.c`.
12* It checks types and expressions in `tree.c`.
13
14To see how a specific lint message is triggered, read the corresponding unit
15test in `tests/usr.bin/xlint/lint1/msg_???.c`.
16
17# Features
18
19## Type checking
20
21Lint has stricter type checking than most C compilers.
22
23In _strict bool mode_, lint treats `bool` as a type that is incompatible with
24other scalar types, like in C#, Go, Java.
25See the test `d_c99_bool_strict.c` for details.
26
27Lint warns about type conversions that may result in alignment problems.
28See the test `msg_135.c` for examples.
29
30## Control flow analysis
31
32Lint roughly tracks the control flow inside a single function.
33It doesn't follow `goto` statements precisely though,
34it rather assumes that each label is reachable.
35See the test `msg_193.c` for examples.
36
37## Error handling
38
39Lint tries to continue parsing and checking even after seeing errors.
40This part of lint is not robust though, so expect some crashes here,
41as variables may not be properly initialized or be null pointers.
42The cleanup after handling a parse error is often incomplete.
43
44## Configurable diagnostic messages
45
46Whether lint prints a message and whether each message is an error, a warning
47or just informational depends on several things:
48
49* The language level, with its possible values:
50    * traditional C (`-t`)
51    * migration from traditional C and C90 (default)
52    * C90 (`-s`)
53    * C99 (`-S`)
54    * C11 (`-Ac11`)
55* In GCC mode (`-g`), lint allows several GNU extensions,
56  reducing the amount of printed messages.
57* In strict bool mode (`-T`), lint issues errors when `bool` is mixed with
58  other scalar types, reusing the existing messages 107 and 211, while also
59  defining new messages that are specific to strict bool mode.
60* The option `-a` performs the check for lossy conversions from large integer
61  types, the option `-aa` extends this check to small integer types as well,
62  reusing the same message ID.
63* The option `-X` suppresses arbitrary messages by their message ID.
64* The option `-q` enables additional queries that are not suitable as regular
65  warnings but may be interesting to look at on a case-by-case basis.
66
67# Fundamental types
68
69Lint mainly analyzes expressions (`tnode_t`), which are formed from operators
70(`op_t`) and their operands (`tnode_t`).
71Each node has a type (`type_t`) and a few other properties.
72
73## type_t
74
75The elementary types are `int`, `_Bool`, `unsigned long`, `pointer` and so on,
76as defined in `tspec_t`.
77
78Actual types like `int`, `const char *` are created by `gettyp(INT)`,
79or by deriving new types from existing types, using `block_derive_pointer`,
80`block_derive_array` and `block_derive_function`.
81(See [below](#memory-management) for the meaning of the prefix `block_`.)
82
83After a type has been created, it should not be modified anymore.
84Ideally all references to types would be `const`, but that's a lot of work.
85Before modifying a type,
86it needs to be copied using `block_dup_type` or `expr_dup_type`.
87
88## tnode_t
89
90When lint parses an expression,
91it builds a tree of nodes representing the AST.
92Each node has an operator that defines which other members may be accessed.
93The operators and their properties are defined in `ops.def`.
94Some examples for operators:
95
96| Operator | Meaning                                                 |
97|----------|---------------------------------------------------------|
98| CON      | compile-time constant in `tn_val`                       |
99| NAME     | references the identifier in `tn_sym`                   |
100| UPLUS    | the unary operator `+tn_left`                           |
101| PLUS     | the binary operator `tn_left + tn_right`                |
102| CALL     | a function call, typically CALL(LOAD(NAME("function"))) |
103| ICALL    | an indirect function call                               |
104| CVT      | an implicit conversion or an explicit cast              |
105
106See `debug_node` for how to interpret the members of `tnode_t`.
107
108## sym_t
109
110There is a single symbol table (`symtab`) for the whole translation unit.
111This means that the same identifier may appear multiple times.
112To distinguish the identifiers, each symbol has a block level.
113Symbols from inner scopes are added to the beginning of the table,
114so they are found first when looking for the identifier.
115
116# Memory management
117
118## Block scope
119
120The memory that is allocated by the `block_*_alloc` functions is freed at the
121end of analyzing the block, that is, after the closing `}`.
122See `compound_statement_rbrace:` in `cgram.y`.
123
124## Expression scope
125
126The memory that is allocated by the `expr_*_alloc` functions is freed at the
127end of analyzing the expression.
128See `expr_free_all`.
129
130# Null pointers
131
132* Expressions can be null.
133    * This typically happens in case of syntax errors or other errors.
134* The subtype of a pointer, array or function is never null.
135
136# Common variable names
137
138| Name | Type      | Meaning                                              |
139|------|-----------|------------------------------------------------------|
140| t    | `tspec_t` | a simple type such as `INT`, `FUNC`, `PTR`           |
141| tp   | `type_t`  | a complete type such as `pointer to array[3] of int` |
142| stp  | `type_t`  | the subtype of a pointer, array or function          |
143| tn   | `tnode_t` | a tree node, mostly used for expressions             |
144| op   | `op_t`    | an operator used in an expression                    |
145| ln   | `tnode_t` | the left-hand operand of a binary operator           |
146| rn   | `tnode_t` | the right-hand operand of a binary operator          |
147| sym  | `sym_t`   | a symbol from the symbol table                       |
148
149# Abbreviations in variable names
150
151| Abbr | Expanded                                    |
152|------|---------------------------------------------|
153| l    | left                                        |
154| r    | right                                       |
155| o    | old (during type conversions)               |
156| n    | new (during type conversions)               |
157| op   | operator                                    |
158| arg  | the number of the argument, for diagnostics |
159
160# Debugging
161
162Useful breakpoints are:
163
164| Function/Code       | File    | Remarks                                              |
165|---------------------|---------|------------------------------------------------------|
166| build_binary        | tree.c  | Creates an expression for a unary or binary operator |
167| initialization_expr | init.c  | Checks a single initializer                          |
168| expr                | tree.c  | Checks a full expression                             |
169| typeok              | tree.c  | Checks two types for compatibility                   |
170| vwarning_at         | err.c   | Prints a warning                                     |
171| verror_at           | err.c   | Prints an error                                      |
172| assert_failed       | err.c   | Prints the location of a failed assertion            |
173| `switch (yyn)`      | cgram.c | Reduction of a grammar rule                          |
174
175# Tests
176
177The tests are in `tests/usr.bin/xlint`.
178By default, each test is run with the lint flags `-g` for GNU mode,
179`-S` for C99 mode and `-w` to report warnings as errors.
180
181Each test can override the lint flags using comments of the following forms:
182
183* `/* lint1-flags: -tw */` replaces the default flags.
184* `/* lint1-extra-flags: -p */` adds to the default flags.
185
186Most tests check the diagnostics that lint generates.
187They do this by placing `expect` comments near the location of the diagnostic.
188The comment `/* expect+1: ... */` expects a diagnostic to be generated for the
189code 1 line below, `/* expect-5: ... */` expects a diagnostic to be generated
190for the code 5 lines above.
191Each `expect` comment must be in a single line.
192At the start and the end of the comment, the placeholder `...` stands for an
193arbitrary sequence of characters.
194There may be other code or comments in the same line of the `.c` file.
195
196Each diagnostic has its own test `msg_???.c` that triggers the corresponding
197diagnostic.
198Most other tests focus on a single feature.
199
200## Adding a new test
201
2021. Run `make add-test NAME=test_name`.
2032. Run `cd ../../../tests/usr.bin/xlint/lint1`.
2043. Sort the `FILES` lines in `Makefile`.
2054. Make the test generate the desired diagnostics.
2065. Run `./accept.sh test_name` until it no longer complains.
2076. Run `cd ../../..`.
2087. Run `cvs commit distrib/sets/lists/tests/mi tests/usr.bin/xlint`.
209