asyn-rs 0.25.0

//! The one C escape table — libCom `epicsString.c`.
//!
//! EPICS escapes a byte in exactly one way, and the callers that print a
//! terminator, a trace line, a record's TINP or an echo mismatch all reach the
//! same table:
//!
//! ```text
//! \a \b \f \n \r \t \v \\ \' \"   the named escapes
//! the byte itself                 when isprint() in the C locale
//! \xNN                            everything else, lower-case hex
//! ```
//!
//! There are two C entry points, and in C they diverge on **the NUL byte**
//! (CBUG-D4): `epicsStrnEscapedFromRaw` has a deliberate `case '\0'` printing
//! `\0` (:145), while `epicsStrPrintEscaped` forgot it, so NUL falls through to
//! the `\xNN` default and prints `\x00` (:259). Under the clean-is-the-goal
//! strategy (`doc/strategy-2026-07-13.md §2`) the port **refuses** that
//! divergence: both forms render NUL as `\0` — C's deliberate escape, the one
//! its round-trippable decoder `epicsStrnRawFromEscaped` was written to accept
//! (`case '0'`). The refusal is structural: [`escape`] holds the single table
//! and there is no per-form NUL parameter to disagree over, so the two renderings
//! are unrepresentable by construction. The C bug was a *missing* named case; the
//! fix supplies it to both, rather than deleting the deliberate one.
//!
//! The two entry points still differ where C genuinely differs — but not on the
//! escape table:
//!
//! - [`escaped_from_raw`] — `epicsStrnEscapedFromRaw` (epicsString.c:120-160)
//!   writes into a *sized* destination. `epicsStrSnPrintEscaped` is a macro alias
//!   for it (epicsString.h:125). This is what `asynShowEos`
//!   (asynShellCommands.c:305), the echo interpose (asynInterposeEcho.c:73-74),
//!   the trace errlog branch (asynManager.c:3159) and every `asynRecord` field
//!   (asynRecord.c:725,1629,2005) escape through — each with its own buffer size,
//!   which the caller must state.
//! - [`print_escaped`] — `epicsStrPrintEscaped` (epicsString.c:230-262), the
//!   `FILE *` form, has no destination bound, and keeps C's `strlen(s) == 0`
//!   first-byte-NUL early-return quirk (R17-49). This is what
//!   `asynPortDriver::report` prints the EOS pair with
//!   (asynPortDriver.cpp:3687,3690).
//!
//! Every escaping caller in this crate names one of the two, so the table exists
//! once. A private per-caller table is how the report came to escape `\r` and
//! `\n` and write a `\x03` terminator raw into stdout (R16-48).

/// C `epicsStrnEscapedFromRaw(dst, dstlen, src, srclen)` (epicsString.c:120-160)
/// — NUL escapes as `\0` (C's deliberate `case '\0'`), and **the destination
/// bound is part of the call**.
///
/// C's `OUT` macro is `ndst++; if (--rem > 0) *dst++ = chr` with `rem = dstlen`:
/// at most `dstlen - 1` escaped characters reach the buffer before the closing
/// NUL, and the cut falls wherever the count runs out — *inside* an escape pair
/// if that is where it lands, leaving a dangling backslash. Compiled libCom
/// (`epicsStrnEscapedFromRaw`, 200 raw CRLF bytes, `dstlen = 40`) returns 400
/// and leaves a 39-char string ending `\r\n\r\`. Every caller therefore states
/// the size of the C buffer it is writing into; there is no unbounded form of
/// this function in C.
pub(crate) fn escaped_from_raw(src: &[u8], dstlen: usize) -> String {
    escape(src, dstlen.saturating_sub(1))
}

/// C `epicsStrPrintEscaped` (epicsString.c:230-262) — the `FILE *` form. C's
/// `switch` forgot the NUL case, so C prints `\x00`; the port **refuses that
/// CBUG-D4 divergence** and renders NUL as `\0`, identically to
/// [`escaped_from_raw`] (both reach the one [`escape`] table). It writes to a
/// stream, so it has no destination bound.
///
/// It takes an explicit `len`, yet guards on `strlen(s) == 0` (`:236-237`):
/// a buffer whose **first byte** is NUL is "no work to do" and prints
/// *nothing*, however many bytes follow it. A NUL anywhere later is escaped
/// normally — `strlen` only looks at the first one. Compiled libCom:
/// `epicsStrPrintEscaped(fp, "\0a", 2)` returns 0 and prints nothing, while
/// `("a\0b", 3)` prints an escaped interior NUL (C `a\x00b`; the port refuses
/// CBUG-D4 and renders `a\0b`). This first-byte-NUL early-return is a C quirk,
/// not a Rust convenience: the ESCAPE trace form on a `FILE *` sink prints an
/// empty data line for a payload that starts with NUL. It is a *separate*
/// behaviour from the NUL-rendering divergence, and is kept.
pub(crate) fn print_escaped(src: &[u8]) -> String {
    if src.first().is_none_or(|&c| c == 0) {
        return String::new();
    }
    escape(src, usize::MAX)
}

/// The shared table. `max` is the number of escaped characters that fit in the
/// destination — C's `dstlen - 1`, or `usize::MAX` for the stream form. The cut
/// is per *character*, not per escape pair, which is what produces C's dangling
/// backslash.
///
/// NUL renders as `\0` for *both* entry points — the CBUG-D4 refusal is
/// structural: there is no NUL parameter for the two forms to disagree over.
fn escape(src: &[u8], max: usize) -> String {
    let mut out = String::with_capacity(src.len().min(max));
    let mut hex = String::new();
    for &c in src {
        let esc = match c {
            0x07 => "\\a",
            0x08 => "\\b",
            0x0c => "\\f",
            b'\n' => "\\n",
            b'\r' => "\\r",
            b'\t' => "\\t",
            0x0b => "\\v",
            b'\\' => "\\\\",
            b'\'' => "\\'",
            b'"' => "\\\"",
            0 => "\\0",
            // C `isprint` in the C locale: the printable ASCII range.
            0x20..=0x7e => {
                hex.clear();
                hex.push(c as char);
                &hex
            }
            _ => {
                use std::fmt::Write;
                hex.clear();
                let _ = write!(hex, "\\x{c:02x}");
                &hex
            }
        };
        for ch in esc.chars() {
            if out.len() >= max {
                return out;
            }
            out.push(ch);
        }
    }
    out
}

#[cfg(test)]
mod tests {
    use super::*;

    /// The two C entry points share one escape table. In C they disagreed on
    /// NUL (`\0` vs `\x00`, CBUG-D4); the port refuses that divergence, so both
    /// forms now render every byte — NUL included — identically.
    #[test]
    fn the_table_is_c_s_and_both_forms_now_agree_on_nul() {
        let named = b"\x07\x08\x0c\n\r\t\x0b\\'\"";
        assert_eq!(escaped_from_raw(named, 80), r#"\a\b\f\n\r\t\v\\\'\""#);
        assert_eq!(print_escaped(named), r#"\a\b\f\n\r\t\v\\\'\""#);

        // isprint → the byte; anything else → lower-case \xNN.
        assert_eq!(escaped_from_raw(b" ~OK", 80), " ~OK");
        assert_eq!(
            escaped_from_raw(b"\x03\x1b\x7f\xff", 80),
            r"\x03\x1b\x7f\xff"
        );
        assert_eq!(print_escaped(b"\x03\x1b\x7f\xff"), r"\x03\x1b\x7f\xff");

        // CBUG-D4 refused: C's `epicsStrnEscapedFromRaw` has a `case '\0'`
        // (epicsString.c:145) and `epicsStrPrintEscaped` does not (:255-260), so
        // C prints `\0` vs `\x00`. The port supplies the missing case to both —
        // NUL is `\0` from either form. Round-trips through the port decoder:
        // `\0` decodes to a single NUL with no octal continuation (C `case '0'`).
        assert_eq!(escaped_from_raw(b"a\0b", 80), r"a\0b");
        assert_eq!(print_escaped(b"a\0b"), r"a\0b");
    }

    /// C's destination bound, byte for byte against compiled libCom
    /// (`epicsStrnEscapedFromRaw` linked from epics-base):
    ///
    /// ```text
    /// 200 CRLF -> 40   ret=400  strlen=39  [\r\n...\r\]
    /// 200 CRLF -> 10   ret=400  strlen= 9  [\r\n\r\n\]
    /// "a\tb"   ->  4   ret=  4  strlen= 3  [a\t]
    /// ```
    ///
    /// The cut is at `dstlen - 1` *characters* of the escaped stream, so it can
    /// fall between the backslash and the letter of an escape pair and leave the
    /// backslash dangling. That is C's output and it is what the record fields
    /// and the trace buffer actually hold.
    #[test]
    fn the_destination_bound_truncates_the_escaped_stream_mid_pair() {
        let crlf: Vec<u8> = b"\r\n".repeat(100);

        let tinp = escaped_from_raw(&crlf, 40);
        assert_eq!(tinp.len(), 39);
        assert!(tinp.ends_with(r"\r\n\r\"), "dangling backslash: {tinp}");

        let eos = escaped_from_raw(&crlf, 10);
        assert_eq!(eos, r"\r\n\r\n\");

        // The pair that does not fit is cut, not dropped whole.
        assert_eq!(escaped_from_raw(b"a\tb", 4), r"a\t");
        assert_eq!(escaped_from_raw(b"a\tb", 3), r"a\");

        // Exactly-fits and under-fits leave the escaped text intact.
        assert_eq!(escaped_from_raw(b"\r\n", 5), r"\r\n");
        assert_eq!(escaped_from_raw(b"", 40), "");

        // A four-character \xNN pair is cut the same way.
        assert_eq!(escaped_from_raw(b"\xff", 3), r"\x");
    }

    /// R17-49. `epicsStrPrintEscaped` early-returns on `strlen(s) == 0`
    /// (epicsString.c:236-237) even though it was handed an explicit length: a
    /// buffer whose first byte is NUL prints nothing at all. Against compiled
    /// libCom:
    ///
    /// ```text
    /// epicsStrPrintEscaped(fp, "\0a", 2)  -> ret 0, prints nothing
    /// epicsStrPrintEscaped(fp, "a\0b", 3) -> ret 6, prints a\x00b in C;
    ///                                        port refuses CBUG-D4 -> a\0b
    /// epicsStrPrintEscaped(fp, "",   0)   -> ret 0, prints nothing
    /// ```
    ///
    /// `epicsStrnEscapedFromRaw` has no such guard — it loops on `srclen` and
    /// escapes the leading NUL as `\0` (compiled: `"\0a"` → `\0a`).
    #[test]
    fn print_escaped_prints_nothing_when_the_first_byte_is_nul() {
        assert_eq!(print_escaped(b"\0a"), "");
        assert_eq!(print_escaped(b"\0"), "");
        assert_eq!(print_escaped(b""), "");

        // Only the FIRST byte: strlen stops there, and a later NUL still escapes
        // — now as `\0` (CBUG-D4 refused), identically to escaped_from_raw.
        assert_eq!(print_escaped(b"a\0b"), r"a\0b");

        // The bounded form keeps escaping — the guard is C's, and C put it in
        // one function only.
        assert_eq!(escaped_from_raw(b"\0a", 40), r"\0a");
    }
}