tidy/
pal.rs

1//! Tidy check to enforce rules about platform-specific code in std.
2//!
3//! This is intended to maintain existing standards of code
4//! organization in hopes that the standard library will continue to
5//! be refactored to isolate platform-specific bits, making porting
6//! easier; where "standard library" roughly means "all the
7//! dependencies of the std and test crates".
8//!
9//! This generally means placing restrictions on where `cfg(unix)`,
10//! `cfg(windows)`, `cfg(target_os)` and `cfg(target_env)` may appear,
11//! the basic objective being to isolate platform-specific code to the
12//! platform-specific `std::sys` modules, and to the allocation,
13//! unwinding, and libc crates.
14//!
15//! Following are the basic rules, though there are currently
16//! exceptions:
17//!
18//! - core may not have platform-specific code.
19//! - libpanic_abort may have platform-specific code.
20//! - libpanic_unwind may have platform-specific code.
21//! - libunwind may have platform-specific code.
22//! - other crates in the std facade may not.
23//! - std may have platform-specific code in the following places:
24//!   - `sys/`
25//!   - `os/`
26//!
27//! `std/sys_common` should _not_ contain platform-specific code.
28//! Finally, because std contains tests with platform-specific
29//! `ignore` attributes, once the parser encounters `mod tests`,
30//! platform-specific cfgs are allowed. Not sure yet how to deal with
31//! this in the long term.
32
33use std::path::Path;
34
35use crate::walk::{filter_dirs, walk};
36
37// Paths that may contain platform-specific code.
38const EXCEPTION_PATHS: &[&str] = &[
39    "library/compiler-builtins",
40    "library/windows_targets",
41    "library/panic_abort",
42    "library/panic_unwind",
43    "library/unwind",
44    "library/rtstartup", // Not sure what to do about this. magic stuff for mingw
45    "library/test",      // Probably should defer to unstable `std::sys` APIs.
46    // The `VaList` implementation must have platform specific code.
47    // The Windows implementation of a `va_list` is always a character
48    // pointer regardless of the target architecture. As a result,
49    // we must use `#[cfg(windows)]` to conditionally compile the
50    // correct `VaList` structure for windows.
51    "library/core/src/ffi/va_list.rs",
52    // core::ffi contains platform-specific type and linkage configuration
53    "library/core/src/ffi/mod.rs",
54    "library/core/src/ffi/primitives.rs",
55    "library/std/src/sys", // Platform-specific code for std lives here.
56    "library/std/src/os",  // Platform-specific public interfaces
57    // Temporary `std` exceptions
58    // FIXME: platform-specific code should be moved to `sys`
59    "library/std/src/io/copy.rs",
60    "library/std/src/io/stdio.rs",
61    "library/std/src/lib.rs", // for miniz_oxide leaking docs, which itself workaround
62    "library/std/src/path.rs",
63    "library/std/src/sys_common", // Should only contain abstractions over platforms
64    "library/std/src/net/test.rs", // Utility helpers for tests
65    "library/std/src/io/error.rs", // Repr unpacked needed for UEFI
66];
67
68pub fn check(path: &Path, bad: &mut bool) {
69    // Sanity check that the complex parsing here works.
70    let mut saw_target_arch = false;
71    let mut saw_cfg_bang = false;
72    walk(path, |path, _is_dir| filter_dirs(path), &mut |entry, contents| {
73        let file = entry.path();
74        let filestr = file.to_string_lossy().replace("\\", "/");
75        if !filestr.ends_with(".rs") {
76            return;
77        }
78
79        let is_exception_path = EXCEPTION_PATHS.iter().any(|s| filestr.contains(&**s));
80        if is_exception_path {
81            return;
82        }
83
84        // exclude tests and benchmarks as some platforms do not support all tests
85        if filestr.contains("tests") || filestr.contains("benches") {
86            return;
87        }
88
89        check_cfgs(contents, &file, bad, &mut saw_target_arch, &mut saw_cfg_bang);
90    });
91
92    assert!(saw_target_arch);
93    assert!(saw_cfg_bang);
94}
95
96fn check_cfgs(
97    contents: &str,
98    file: &Path,
99    bad: &mut bool,
100    saw_target_arch: &mut bool,
101    saw_cfg_bang: &mut bool,
102) {
103    // Pull out all `cfg(...)` and `cfg!(...)` strings.
104    let cfgs = parse_cfgs(contents);
105
106    let mut line_numbers: Option<Vec<usize>> = None;
107    let mut err = |idx: usize, cfg: &str| {
108        if line_numbers.is_none() {
109            line_numbers = Some(contents.match_indices('\n').map(|(i, _)| i).collect());
110        }
111        let line_numbers = line_numbers.as_ref().expect("");
112        let line = match line_numbers.binary_search(&idx) {
113            Ok(_) => unreachable!(),
114            Err(i) => i + 1,
115        };
116        tidy_error!(bad, "{}:{}: platform-specific cfg: {}", file.display(), line, cfg);
117    };
118
119    for (idx, cfg) in cfgs {
120        // Sanity check that the parsing here works.
121        if !*saw_target_arch && cfg.contains("target_arch") {
122            *saw_target_arch = true
123        }
124        if !*saw_cfg_bang && cfg.contains("cfg!") {
125            *saw_cfg_bang = true
126        }
127
128        let contains_platform_specific_cfg = cfg.contains("target_os")
129            || cfg.contains("target_env")
130            || cfg.contains("target_abi")
131            || cfg.contains("target_vendor")
132            || cfg.contains("target_family")
133            || cfg.contains("unix")
134            || cfg.contains("windows");
135
136        if !contains_platform_specific_cfg {
137            continue;
138        }
139
140        let preceded_by_doc_comment = {
141            let pre_contents = &contents[..idx];
142            let pre_newline = pre_contents.rfind('\n');
143            let pre_doc_comment = pre_contents.rfind("///");
144            match (pre_newline, pre_doc_comment) {
145                (Some(n), Some(c)) => n < c,
146                (None, Some(_)) => true,
147                (_, None) => false,
148            }
149        };
150
151        if preceded_by_doc_comment {
152            continue;
153        }
154
155        // exclude tests as some platforms do not support all tests
156        if cfg.contains("test") {
157            continue;
158        }
159
160        err(idx, cfg);
161    }
162}
163
164fn parse_cfgs(contents: &str) -> Vec<(usize, &str)> {
165    let candidate_cfgs = contents.match_indices("cfg");
166    let candidate_cfg_idxs = candidate_cfgs.map(|(i, _)| i);
167    // This is puling out the indexes of all "cfg" strings
168    // that appear to be tokens followed by a parenthesis.
169    let cfgs = candidate_cfg_idxs.filter(|i| {
170        let pre_idx = i.saturating_sub(1);
171        let succeeds_non_ident = !contents
172            .as_bytes()
173            .get(pre_idx)
174            .cloned()
175            .map(char::from)
176            .map(char::is_alphanumeric)
177            .unwrap_or(false);
178        let contents_after = &contents[*i..];
179        let first_paren = contents_after.find('(');
180        let paren_idx = first_paren.map(|ip| i + ip);
181        let preceeds_whitespace_and_paren = paren_idx
182            .map(|ip| {
183                let maybe_space = &contents[*i + "cfg".len()..ip];
184                maybe_space.chars().all(|c| char::is_whitespace(c) || c == '!')
185            })
186            .unwrap_or(false);
187
188        succeeds_non_ident && preceeds_whitespace_and_paren
189    });
190
191    cfgs.flat_map(|i| {
192        let mut depth = 0;
193        let contents_from = &contents[i..];
194        for (j, byte) in contents_from.bytes().enumerate() {
195            match byte {
196                b'(' => {
197                    depth += 1;
198                }
199                b')' => {
200                    depth -= 1;
201                    if depth == 0 {
202                        return Some((i, &contents_from[..=j]));
203                    }
204                }
205                _ => {}
206            }
207        }
208
209        // if the parentheses are unbalanced just ignore this cfg -- it'll be caught when attempting
210        // to run the compiler, and there's no real reason to lint it separately here
211        None
212    })
213    .collect()
214}