Add explicit Bash redirect analysis

Author: Aaronontheweb

IMPLEMENTATION_PLAN.md

@@ -305,6 +305,21 @@ priorities.
       transfers, and occurrence-specific redirect values remain fail closed.
       Next add the Netclaw approval matrix before calling the Bash consumer
       integration complete.
+- [x] Deliver occurrence-level Bash explicit redirect facts for ordinary file
+      input/output/append, static descriptor duplicate/close/move, computed
+      descriptor targets, and combined `&>` / `&>>` output. Static descriptor
+      operations are complete non-path facts even though the v0.2 compatibility
+      redirect remains `IsDynamicSkip`; computed targets remain incomplete and
+      cannot be exempted by raw prefix. Arbitrary numeric source descriptors are
+      retained only when authored at a token boundary; overflow sources remain
+      incomplete rather than being truncated, and word suffixes such as
+      `command3>file` keep `3` in the command name. Unquoted LF/CRLF line
+      continuations are removed before descriptor recognition, including when
+      they join multi-digit sources. Exact file targets now complete their
+      containing occurrence, while cwd or value uncertainty still downgrades
+      the redirect and occurrence after abstract-state joins. Direct lexer and
+      parser tests plus executable corpus cases pin the boundary. Next map the
+      PowerShell stream model, then prove the paired Netclaw redirect matrix.
 - [ ] Complete PowerShell `foreach` integration and add the
       Netclaw approval-matrix cases. The structural slice now preserves literal
       scalar/array and executable iterator forms, recursively parses bodies,

SPEC.md

@@ -967,8 +967,9 @@ public sealed record Clause
     public IReadOnlyList<Arg> Args { get; init; } = [];
 
     /// <summary>
-    /// Redirect operators on this clause (>, >>, <, 2>, 2>>). Each entry
-    /// includes direction and target path.
+    /// Compatibility redirects on this clause (>, >>, <, 2>, 2>>, &>,
+    /// &>>). Each entry retains the v0.2 direction and target projection;
+    /// v0.3 consumers use CommandOccurrence.Redirects for exact semantics.
     /// </summary>
     public IReadOnlyList<Redirect> Redirects { get; init; } = [];
 
@@ -1270,7 +1271,8 @@ verb_like_word  := static word satisfying §6.1; the initial command-name
 arg             := word | flag | quoted_string | supported_substitution
 flag            := "-" letter+ | "--" word
 redirect        := redirect_op target
-redirect_op     := ">" | ">>" | "<" | "2>" | "2>>"
+redirect_op     := descriptor? (">" | ">>" | "<") | "&>" | "&>>"
+descriptor      := digit+
 target          := word | quoted_string | supported_substitution
 supported_substitution := "$(" command ")"
 word            := non-whitespace, non-operator fragments; may contain
@@ -1396,7 +1398,8 @@ The lexer produces tokens consumed by the parser. Token kinds:
 - **QUOTED_STRING** — single- or double-quoted string. The lexer strips
   the quote delimiters from the token value. Example: `"hello world"`
   becomes the token value `hello world`.
-- **OPERATOR** — `&&`, `||`, `;`, `|`, `>`, `>>`, `<`, `2>`, `2>>`,
+- **OPERATOR** — `&&`, `||`, `;`, `|`, `>`, `>>`, `<`, numeric-descriptor
+  forms such as `2>`, `3>>`, and `10<`, `&>`, `&>>`,
   `(`, `)`, `<<`, `<<-`.
 - **WHITESPACE** — one or more spaces, tabs, or newlines (newlines inside
   a heredoc body are not emitted as ordinary tokens; the delimiter token
@@ -1456,15 +1459,22 @@ The lexer produces tokens consumed by the parser. Token kinds:
 
 Operators terminate the current token. `cd /tmp&&ls` lexes as
 `[cd, /tmp, &&, ls]` — no whitespace required around operators. The lexer
-must handle this.
+must handle this. A numeric descriptor is an operator prefix only when its
+digits begin at a shell-token boundary and become adjacent to `<`, `>`, or
+`>>` after Bash removes unquoted line continuations. Continuations may join
+digit fragments or the descriptor and operator; LF and CRLF spellings retain
+their authored span while producing the same descriptor. Digits joined to an
+ordinary, quoted, or escaped word remain part of that word; `command3>file`
+therefore uses command name `command3` and a default-source `>` redirect.
 
 ### Comment handling
 
 - An unquoted `#` that appears at a **word boundary** starts a comment
   that runs to (but does not include) the next newline. A word boundary
   is: start of input, or the position immediately after a whitespace
   run, a newline, an operator (`&&`, `||`, `;`, `|`, `>`, `>>`, `<`,
-  `2>`, `2>>`, `(`, `)`, `<<`, `<<-`), a quoted string, or an opaque
+  a numeric descriptor adjacent to `>`, `>>`, or `<`, `&>`, `&>>`, `(`,
+  `)`, `<<`, `<<-`), a quoted string, or an opaque
   substitution. Equivalently: `#` is comment-start everywhere the
   outer lexer dispatch loop sits, because every other lexer rule has
   already consumed its territory before `#` is considered.

openspec/changes/v0-3-structured-shell-analysis/tasks.md

@@ -65,11 +65,20 @@
 ## 4. Explicit Redirect Semantics
 
 - [x] 4.1 Add the locked redirect operation and target-analysis types while retaining compatibility redirect members.
-- [ ] 4.2 Classify Bash descriptor duplication, close, and move as static only for the complete literal descriptor grammar.
-- [ ] 4.3 Keep variable-driven and otherwise computed Bash descriptor targets unknown or incomplete.
-- [ ] 4.4 Lex and classify Bash `&>` and `&>>` independently from background-list operators.
+- [x] 4.2 Classify Bash descriptor duplication, close, and move as static only for the complete literal descriptor grammar.
+- [x] 4.3 Keep variable-driven and otherwise computed Bash descriptor targets unknown or incomplete.
+- [x] 4.4 Lex and classify Bash `&>` and `&>>` independently from background-list operators.
 - [ ] 4.5 Map existing PowerShell stream redirects into the shared explicit model without losing shell-specific stream identity.
 - [ ] 4.6 Add paired direct tests and corpus cases for static, dynamic, malformed, multiple, combined, and file redirects.
+  - [x] 4.6a Add Bash direct and executable-corpus cases for static duplicate,
+    close, and move; computed descriptor targets; combined output overwrite and
+    append; ordinary file redirects; arbitrary numeric source descriptors; and
+    overflow/malformed fail-closed boundaries. LF/CRLF continuations at the
+    descriptor boundary and inside multi-digit sources follow Bash's pre-token
+    removal semantics. Executable-corpus cases pin malformed atomic failure and
+    independent occurrence facts for multiple redirects.
+  - [ ] 4.6b Add the paired PowerShell direct and executable-corpus cases when
+    task 4.5 maps its stream model.
 - [ ] 4.7 Verify the explicit model removes the need for raw-prefix inference in a Netclaw integration test.
 
 ## 5. Consumer Migration Baseline

src/ShellSyntaxTree/Internal/Bash/Lexing/BashLexer.cs

@@ -114,10 +114,16 @@ internal static IReadOnlyList<BashToken> Tokenize(string input)
             }
 
             // ---- operators (longer-match first) ----
-            // Order matters: `&&` before `&`, `||` before `|`, `>>` before `>`,
-            // `2>>` before `2>`, `<<-` before `<<`, `<<` before `<`. We don't
-            // recognize a bare `&` in v0.1 (no background-job support; SPEC §1).
-            if (TryReadOperator(src, i, out var opLen, out var opText))
+            // Order matters: a token-boundary numeric descriptor precedes its
+            // redirect, `&&` precedes `&`, `||` precedes `|`, `>>` precedes
+            // `>`, and `<<-` precedes `<<` and `<`. Bare `&` background jobs
+            // remain unsupported.
+            if (TryReadOperator(
+                    src,
+                    i,
+                    CanStartNumericDescriptor(tokens),
+                    out var opLen,
+                    out var opText))
             {
                 var operatorTok = new BashToken(
                     BashTokenKind.Operator, "", opText, i, opLen, null);
@@ -268,7 +274,7 @@ private static bool TryConsumeFileDescriptorTarget(
         }
 
         if (previousIndex < 0 || tokens[previousIndex].Kind != BashTokenKind.Operator ||
-            tokens[previousIndex].OperatorText is not (">" or ">>" or "<" or "2>" or "2>>"))
+            !CanTakeDescriptorTarget(tokens[previousIndex].OperatorText))
         {
             return false;
         }
@@ -323,16 +329,89 @@ tokens[previousIndex].OperatorText is not (">" or ">>" or "<" or "2>" or "2>>"))
         return true;
     }
 
+    private static bool CanTakeDescriptorTarget(string? operatorText)
+    {
+        if (string.IsNullOrEmpty(operatorText))
+        {
+            return false;
+        }
+
+        var operatorStart = 0;
+        while (operatorStart < operatorText!.Length &&
+               operatorText[operatorStart] is >= '0' and <= '9')
+        {
+            operatorStart++;
+        }
+
+        var redirect = operatorText.Substring(operatorStart);
+        return redirect is ">" or ">>" or "<";
+    }
+
     // ---------------------------------------------------------------- operators
 
     private static bool TryReadOperator(
-        ReadOnlySpan<char> src, int i, out int length, out string? text)
+        ReadOnlySpan<char> src,
+        int i,
+        bool canStartNumericDescriptor,
+        out int length,
+        out string? text)
     {
+        var descriptor = new StringBuilder();
+        var descriptorEnd = i;
+        while (descriptorEnd < src.Length)
+        {
+            if (src[descriptorEnd] is >= '0' and <= '9')
+            {
+                descriptor.Append(src[descriptorEnd]);
+                descriptorEnd++;
+                continue;
+            }
+
+            if (TrySkipLineContinuation(src, descriptorEnd, out var afterContinuation))
+            {
+                descriptorEnd = afterContinuation;
+                continue;
+            }
+
+            break;
+        }
+
+        if (canStartNumericDescriptor &&
+            descriptor.Length > 0 &&
+            descriptorEnd < src.Length)
+        {
+            if (src[descriptorEnd] == '>')
+            {
+                var append = descriptorEnd + 1 < src.Length &&
+                    src[descriptorEnd + 1] == '>';
+                length = descriptorEnd - i + (append ? 2 : 1);
+                text = descriptor.ToString() + (append ? ">>" : ">");
+                return true;
+            }
+
+            if (src[descriptorEnd] == '<')
+            {
+                length = descriptorEnd - i + 1;
+                text = descriptor.ToString() + "<";
+                return true;
+            }
+        }
+
         // Multi-char operators first.
         if (i + 1 < src.Length)
         {
             var c0 = src[i];
             var c1 = src[i + 1];
+            if (c0 == '&' && c1 == '>')
+            {
+                if (i + 2 < src.Length && src[i + 2] == '>')
+                {
+                    length = 3; text = "&>>"; return true;
+                }
+
+                length = 2; text = "&>"; return true;
+            }
+
             if (c0 == '&' && c1 == '&') { length = 2; text = "&&"; return true; }
             if (c0 == '|' && c1 == '|') { length = 2; text = "||"; return true; }
             if (c0 == '>' && c1 == '>') { length = 2; text = ">>"; return true; }
@@ -346,15 +425,6 @@ private static bool TryReadOperator(
                 length = 2; text = "<<"; return true;
             }
 
-            if (c0 == '2' && c1 == '>')
-            {
-                if (i + 2 < src.Length && src[i + 2] == '>')
-                {
-                    length = 3; text = "2>>"; return true;
-                }
-
-                length = 2; text = "2>"; return true;
-            }
         }
 
         // Single-char operators.
@@ -371,6 +441,49 @@ private static bool TryReadOperator(
         }
     }
 
+    private static bool TrySkipLineContinuation(
+        ReadOnlySpan<char> source,
+        int index,
+        out int afterContinuation)
+    {
+        afterContinuation = index;
+        if (index + 1 >= source.Length || source[index] != '\\' ||
+            source[index + 1] is not ('\n' or '\r'))
+        {
+            return false;
+        }
+
+        afterContinuation = source[index + 1] == '\r' &&
+            index + 2 < source.Length && source[index + 2] == '\n'
+            ? index + 3
+            : index + 2;
+        return true;
+    }
+
+    private static bool CanStartNumericDescriptor(IReadOnlyList<BashToken> tokens)
+    {
+        if (tokens.Count == 0)
+        {
+            return true;
+        }
+
+        var index = tokens.Count - 1;
+        while (index >= 0 && tokens[index].Kind == BashTokenKind.Continuation)
+        {
+            index--;
+        }
+
+        if (index < 0)
+        {
+            return true;
+        }
+
+        return tokens[index].Kind is
+            BashTokenKind.Whitespace or
+            BashTokenKind.Operator or
+            BashTokenKind.Comment;
+    }
+
     // ---------------------------------------------------------------- quoted
 
     private static int ReadSingleQuoted(
@@ -1219,10 +1332,6 @@ private static bool IsOperatorStart(ReadOnlySpan<char> src, int i)
                 // Both `&&` and unsupported bare `&` terminate a word. The
                 // tokenizer emits a sentinel for the latter on its next pass.
                 return true;
-            case '2':
-                // `2>` and `2>>` start with '2' — only treat them as operator
-                // starts when the immediate next char is '>'.
-                return i + 1 < src.Length && src[i + 1] == '>';
             default:
                 return false;
         }

src/ShellSyntaxTree/Internal/Bash/Parsing/BashAbstractStateAnalyzer.cs

@@ -1207,14 +1207,15 @@ private SimpleCommandSyntax RewriteSimple(
         }
 
         var clause = RewriteClause(simple.Clause, input, sourceFacts.CwdPathDependencies);
+        var redirects = RewriteRedirectFacts(sourceFacts.Redirects, clause);
         facts.Add(clause, new CommandOccurrenceFacts
         {
             EffectiveArguments = CreateEffectiveArguments(simple.Clause),
             WorkingDirectory = input.ToDomain(),
-            Redirects = RewriteRedirectFacts(sourceFacts.Redirects, clause),
+            Redirects = redirects,
             CwdPathDependencies = sourceFacts.CwdPathDependencies,
             ValueProvenance = sourceFacts.ValueProvenance,
-            IsComplete = sourceFacts.IsComplete,
+            IsComplete = sourceFacts.IsComplete && AreRedirectsComplete(redirects),
         });
         return simple with
         {
@@ -1463,12 +1464,26 @@ private static IReadOnlyList<RedirectAnalysis> RewriteRedirectFacts(
                         Kind = ShellValueDomainKind.Exact,
                         Values = new[] { redirect.Target },
                     },
+                IsComplete = fact.IsComplete && !redirect.IsDynamicSkip,
             };
         }
 
         return rewritten;
     }
 
+    private static bool AreRedirectsComplete(IReadOnlyList<RedirectAnalysis> redirects)
+    {
+        foreach (var redirect in redirects)
+        {
+            if (!redirect.IsComplete)
+            {
+                return false;
+            }
+        }
+
+        return true;
+    }
+
     private string? RebaseResolution(
         string? resolved,
         bool isPath,