Rollup merge of rust-lang#145208 - joshtriplett:mbe-derive, r=petrochenkov

Implement declarative (`macro_rules!`) derive macros (RFC 3698)

This is a draft for review, and should not be merged yet.

This is layered atop rust-lang#145153 , and has
only two additional commits atop that. The first handles parsing and provides a
test for various parse errors. The second implements expansion and handles
application.

This implements RFC 3698, "Declarative (`macro_rules!`) derive macros".
Tracking issue: rust-lang#143549

This has one remaining issue, which I could use some help debugging: in
`tests/ui/macros/macro-rules-derive-error.rs`, the diagnostics for
`derive(fn_only)` (for a `fn_only` with no `derive` rules) and
`derive(ForwardReferencedDerive)` both get emitted twice, as a duplicate
diagnostic.

From what I can tell via adding some debugging code,
`unresolved_macro_suggestions` is getting called twice from
`finalize_macro_resolutions` for each of them, because
`self.single_segment_macro_resolutions` has two entries for the macro, with two
different `parent_scope` values. I'm not clear on why that happened; it doesn't
happen with the equivalent code using attrs.

I'd welcome any suggestions for fixing this.

Author: Zalathar

compiler/rustc_expand/messages.ftl

@@ -70,7 +70,7 @@ expand_invalid_fragment_specifier =
     invalid fragment specifier `{$fragment}`
     .help = {$help}
 
-expand_macro_args_bad_delim = macro attribute argument matchers require parentheses
+expand_macro_args_bad_delim = `{$rule_kw}` rule argument matchers require parentheses
 expand_macro_args_bad_delim_sugg = the delimiters should be `(` and `)`
 
 expand_macro_body_stability =

compiler/rustc_expand/src/errors.rs

@@ -490,6 +490,7 @@ pub(crate) struct MacroArgsBadDelim {
     pub span: Span,
     #[subdiagnostic]
     pub sugg: MacroArgsBadDelimSugg,
+    pub rule_kw: Symbol,
 }
 
 #[derive(Subdiagnostic)]

compiler/rustc_expand/src/expand.rs

@@ -16,6 +16,7 @@ use rustc_attr_parsing::{EvalConfigResult, ShouldEmit};
 use rustc_data_structures::flat_map_in_place::FlatMapInPlace;
 use rustc_errors::PResult;
 use rustc_feature::Features;
+use rustc_hir::def::MacroKinds;
 use rustc_parse::parser::{
     AttemptLocalParseRecovery, CommaRecoveryMode, ForceCollect, Parser, RecoverColon, RecoverComma,
     token_descr,
@@ -565,6 +566,7 @@ impl<'a, 'b> MacroExpander<'a, 'b> {
                                 .map(|DeriveResolution { path, item, exts: _, is_const }| {
                                     // FIXME: Consider using the derive resolutions (`_exts`)
                                     // instead of enqueuing the derives to be resolved again later.
+                                    // Note that this can result in duplicate diagnostics.
                                     let expn_id = LocalExpnId::fresh_empty();
                                     derive_invocations.push((
                                         Invocation {
@@ -922,6 +924,35 @@ impl<'a, 'b> MacroExpander<'a, 'b> {
                     }
                     fragment
                 }
+                SyntaxExtensionKind::MacroRules(expander)
+                    if expander.kinds().contains(MacroKinds::DERIVE) =>
+                {
+                    if is_const {
+                        let guar = self
+                            .cx
+                            .dcx()
+                            .span_err(span, "macro `derive` does not support const derives");
+                        return ExpandResult::Ready(fragment_kind.dummy(span, guar));
+                    }
+                    let body = item.to_tokens();
+                    match expander.expand_derive(self.cx, span, &body) {
+                        Ok(tok_result) => {
+                            let fragment =
+                                self.parse_ast_fragment(tok_result, fragment_kind, &path, span);
+                            if macro_stats {
+                                update_derive_macro_stats(
+                                    self.cx,
+                                    fragment_kind,
+                                    span,
+                                    &path,
+                                    &fragment,
+                                );
+                            }
+                            fragment
+                        }
+                        Err(guar) => return ExpandResult::Ready(fragment_kind.dummy(span, guar)),
+                    }
+                }
                 _ => unreachable!(),
             },
             InvocationKind::GlobDelegation { item, of_trait } => {

compiler/rustc_expand/src/mbe/diagnostics.rs

@@ -14,14 +14,22 @@ use super::macro_rules::{MacroRule, NoopTracker, parser_from_cx};
 use crate::expand::{AstFragmentKind, parse_ast_fragment};
 use crate::mbe::macro_parser::ParseResult::*;
 use crate::mbe::macro_parser::{MatcherLoc, NamedParseResult, TtParser};
-use crate::mbe::macro_rules::{Tracker, try_match_macro, try_match_macro_attr};
+use crate::mbe::macro_rules::{
+    Tracker, try_match_macro, try_match_macro_attr, try_match_macro_derive,
+};
+
+pub(super) enum FailedMacro<'a> {
+    Func,
+    Attr(&'a TokenStream),
+    Derive,
+}
 
 pub(super) fn failed_to_match_macro(
     psess: &ParseSess,
     sp: Span,
     def_span: Span,
     name: Ident,
-    attr_args: Option<&TokenStream>,
+    args: FailedMacro<'_>,
     body: &TokenStream,
     rules: &[MacroRule],
 ) -> (Span, ErrorGuaranteed) {
@@ -36,10 +44,12 @@ pub(super) fn failed_to_match_macro(
     // diagnostics.
     let mut tracker = CollectTrackerAndEmitter::new(psess.dcx(), sp);
 
-    let try_success_result = if let Some(attr_args) = attr_args {
-        try_match_macro_attr(psess, name, attr_args, body, rules, &mut tracker)
-    } else {
-        try_match_macro(psess, name, body, rules, &mut tracker)
+    let try_success_result = match args {
+        FailedMacro::Func => try_match_macro(psess, name, body, rules, &mut tracker),
+        FailedMacro::Attr(attr_args) => {
+            try_match_macro_attr(psess, name, attr_args, body, rules, &mut tracker)
+        }
+        FailedMacro::Derive => try_match_macro_derive(psess, name, body, rules, &mut tracker),
     };
 
     if try_success_result.is_ok() {
@@ -90,7 +100,7 @@ pub(super) fn failed_to_match_macro(
     }
 
     // Check whether there's a missing comma in this macro call, like `println!("{}" a);`
-    if attr_args.is_none()
+    if let FailedMacro::Func = args
         && let Some((body, comma_span)) = body.add_comma()
     {
         for rule in rules {

compiler/rustc_expand/src/mbe/macro_rules.rs

@@ -27,10 +27,10 @@ use rustc_session::Session;
 use rustc_session::parse::{ParseSess, feature_err};
 use rustc_span::edition::Edition;
 use rustc_span::hygiene::Transparency;
-use rustc_span::{Ident, Span, kw, sym};
+use rustc_span::{Ident, Span, Symbol, kw, sym};
 use tracing::{debug, instrument, trace, trace_span};
 
-use super::diagnostics::failed_to_match_macro;
+use super::diagnostics::{FailedMacro, failed_to_match_macro};
 use super::macro_parser::{NamedMatches, NamedParseResult};
 use super::{SequenceRepetition, diagnostics};
 use crate::base::{
@@ -138,6 +138,8 @@ pub(super) enum MacroRule {
         body_span: Span,
         rhs: mbe::TokenTree,
     },
+    /// A derive rule, for use with `#[m]`
+    Derive { body: Vec<MatcherLoc>, body_span: Span, rhs: mbe::TokenTree },
 }
 
 pub struct MacroRulesMacroExpander {
@@ -157,13 +159,71 @@ impl MacroRulesMacroExpander {
             MacroRule::Attr { args_span, body_span, ref rhs, .. } => {
                 (MultiSpan::from_spans(vec![args_span, body_span]), rhs)
             }
+            MacroRule::Derive { body_span, ref rhs, .. } => (MultiSpan::from_span(body_span), rhs),
         };
         if has_compile_error_macro(rhs) { None } else { Some((&self.name, span)) }
     }
 
     pub fn kinds(&self) -> MacroKinds {
         self.kinds
     }
+
+    pub fn expand_derive(
+        &self,
+        cx: &mut ExtCtxt<'_>,
+        sp: Span,
+        body: &TokenStream,
+    ) -> Result<TokenStream, ErrorGuaranteed> {
+        // This is similar to `expand_macro`, but they have very different signatures, and will
+        // diverge further once derives support arguments.
+        let Self { name, ref rules, node_id, .. } = *self;
+        let psess = &cx.sess.psess;
+
+        if cx.trace_macros() {
+            let msg = format!("expanding `#[derive({name})] {}`", pprust::tts_to_string(body));
+            trace_macros_note(&mut cx.expansions, sp, msg);
+        }
+
+        match try_match_macro_derive(psess, name, body, rules, &mut NoopTracker) {
+            Ok((rule_index, rule, named_matches)) => {
+                let MacroRule::Derive { rhs, .. } = rule else {
+                    panic!("try_match_macro_derive returned non-derive rule");
+                };
+                let mbe::TokenTree::Delimited(rhs_span, _, rhs) = rhs else {
+                    cx.dcx().span_bug(sp, "malformed macro derive rhs");
+                };
+
+                let id = cx.current_expansion.id;
+                let tts = transcribe(psess, &named_matches, rhs, *rhs_span, self.transparency, id)
+                    .map_err(|e| e.emit())?;
+
+                if cx.trace_macros() {
+                    let msg = format!("to `{}`", pprust::tts_to_string(&tts));
+                    trace_macros_note(&mut cx.expansions, sp, msg);
+                }
+
+                if is_defined_in_current_crate(node_id) {
+                    cx.resolver.record_macro_rule_usage(node_id, rule_index);
+                }
+
+                Ok(tts)
+            }
+            Err(CanRetry::No(guar)) => Err(guar),
+            Err(CanRetry::Yes) => {
+                let (_, guar) = failed_to_match_macro(
+                    cx.psess(),
+                    sp,
+                    self.span,
+                    name,
+                    FailedMacro::Derive,
+                    body,
+                    rules,
+                );
+                cx.macro_error_and_trace_macros_diag();
+                Err(guar)
+            }
+        }
+    }
 }
 
 impl TTMacroExpander for MacroRulesMacroExpander {
@@ -325,8 +385,15 @@ fn expand_macro<'cx>(
         }
         Err(CanRetry::Yes) => {
             // Retry and emit a better error.
-            let (span, guar) =
-                failed_to_match_macro(cx.psess(), sp, def_span, name, None, &arg, rules);
+            let (span, guar) = failed_to_match_macro(
+                cx.psess(),
+                sp,
+                def_span,
+                name,
+                FailedMacro::Func,
+                &arg,
+                rules,
+            );
             cx.macro_error_and_trace_macros_diag();
             DummyResult::any(span, guar)
         }
@@ -388,8 +455,15 @@ fn expand_macro_attr(
         Err(CanRetry::No(guar)) => Err(guar),
         Err(CanRetry::Yes) => {
             // Retry and emit a better error.
-            let (_, guar) =
-                failed_to_match_macro(cx.psess(), sp, def_span, name, Some(&args), &body, rules);
+            let (_, guar) = failed_to_match_macro(
+                cx.psess(),
+                sp,
+                def_span,
+                name,
+                FailedMacro::Attr(&args),
+                &body,
+                rules,
+            );
             cx.trace_macros_diag();
             Err(guar)
         }
@@ -537,6 +611,44 @@ pub(super) fn try_match_macro_attr<'matcher, T: Tracker<'matcher>>(
     Err(CanRetry::Yes)
 }
 
+/// Try expanding the macro derive. Returns the index of the successful arm and its
+/// named_matches if it was successful, and nothing if it failed. On failure, it's the caller's job
+/// to use `track` accordingly to record all errors correctly.
+#[instrument(level = "debug", skip(psess, body, rules, track), fields(tracking = %T::description()))]
+pub(super) fn try_match_macro_derive<'matcher, T: Tracker<'matcher>>(
+    psess: &ParseSess,
+    name: Ident,
+    body: &TokenStream,
+    rules: &'matcher [MacroRule],
+    track: &mut T,
+) -> Result<(usize, &'matcher MacroRule, NamedMatches), CanRetry> {
+    // This uses the same strategy as `try_match_macro`
+    let body_parser = parser_from_cx(psess, body.clone(), T::recovery());
+    let mut tt_parser = TtParser::new(name);
+    for (i, rule) in rules.iter().enumerate() {
+        let MacroRule::Derive { body, .. } = rule else { continue };
+
+        let mut gated_spans_snapshot = mem::take(&mut *psess.gated_spans.spans.borrow_mut());
+
+        let result = tt_parser.parse_tt(&mut Cow::Borrowed(&body_parser), body, track);
+        track.after_arm(true, &result);
+
+        match result {
+            Success(named_matches) => {
+                psess.gated_spans.merge(gated_spans_snapshot);
+                return Ok((i, rule, named_matches));
+            }
+            Failure(_) => {
+                mem::swap(&mut gated_spans_snapshot, &mut psess.gated_spans.spans.borrow_mut())
+            }
+            Error(_, _) => return Err(CanRetry::Yes),
+            ErrorReported(guar) => return Err(CanRetry::No(guar)),
+        }
+    }
+
+    Err(CanRetry::Yes)
+}
+
 /// Converts a macro item into a syntax extension.
 pub fn compile_declarative_macro(
     sess: &Session,
@@ -570,7 +682,7 @@ pub fn compile_declarative_macro(
     let mut rules = Vec::new();
 
     while p.token != token::Eof {
-        let args = if p.eat_keyword_noexpect(sym::attr) {
+        let (args, is_derive) = if p.eat_keyword_noexpect(sym::attr) {
             kinds |= MacroKinds::ATTR;
             if !features.macro_attr() {
                 feature_err(sess, sym::macro_attr, span, "`macro_rules!` attributes are unstable")
@@ -580,16 +692,46 @@ pub fn compile_declarative_macro(
                 return dummy_syn_ext(guar);
             }
             let args = p.parse_token_tree();
-            check_args_parens(sess, &args);
+            check_args_parens(sess, sym::attr, &args);
             let args = parse_one_tt(args, RulePart::Pattern, sess, node_id, features, edition);
             check_emission(check_lhs(sess, node_id, &args));
             if let Some(guar) = check_no_eof(sess, &p, "expected macro attr body") {
                 return dummy_syn_ext(guar);
             }
-            Some(args)
+            (Some(args), false)
+        } else if p.eat_keyword_noexpect(sym::derive) {
+            kinds |= MacroKinds::DERIVE;
+            let derive_keyword_span = p.prev_token.span;
+            if !features.macro_derive() {
+                feature_err(sess, sym::macro_attr, span, "`macro_rules!` derives are unstable")
+                    .emit();
+            }
+            if let Some(guar) = check_no_eof(sess, &p, "expected `()` after `derive`") {
+                return dummy_syn_ext(guar);
+            }
+            let args = p.parse_token_tree();
+            check_args_parens(sess, sym::derive, &args);
+            let args_empty_result = check_args_empty(sess, &args);
+            let args_not_empty = args_empty_result.is_err();
+            check_emission(args_empty_result);
+            if let Some(guar) = check_no_eof(sess, &p, "expected macro derive body") {
+                return dummy_syn_ext(guar);
+            }
+            // If the user has `=>` right after the `()`, they might have forgotten the empty
+            // parentheses.
+            if p.token == token::FatArrow {
+                let mut err = sess
+                    .dcx()
+                    .struct_span_err(p.token.span, "expected macro derive body, got `=>`");
+                if args_not_empty {
+                    err.span_label(derive_keyword_span, "need `()` after this `derive`");
+                }
+                return dummy_syn_ext(err.emit());
+            }
+            (None, true)
         } else {
             kinds |= MacroKinds::BANG;
-            None
+            (None, false)
         };
         let lhs_tt = p.parse_token_tree();
         let lhs_tt = parse_one_tt(lhs_tt, RulePart::Pattern, sess, node_id, features, edition);
@@ -620,6 +762,8 @@ pub fn compile_declarative_macro(
             let args = mbe::macro_parser::compute_locs(&delimited.tts);
             let body_span = lhs_span;
             rules.push(MacroRule::Attr { args, args_span, body: lhs, body_span, rhs: rhs_tt });
+        } else if is_derive {
+            rules.push(MacroRule::Derive { body: lhs, body_span: lhs_span, rhs: rhs_tt });
         } else {
             rules.push(MacroRule::Func { lhs, lhs_span, rhs: rhs_tt });
         }
@@ -666,18 +810,29 @@ fn check_no_eof(sess: &Session, p: &Parser<'_>, msg: &'static str) -> Option<Err
     None
 }
 
-fn check_args_parens(sess: &Session, args: &tokenstream::TokenTree) {
+fn check_args_parens(sess: &Session, rule_kw: Symbol, args: &tokenstream::TokenTree) {
     // This does not handle the non-delimited case; that gets handled separately by `check_lhs`.
     if let tokenstream::TokenTree::Delimited(dspan, _, delim, _) = args
         && *delim != Delimiter::Parenthesis
     {
         sess.dcx().emit_err(errors::MacroArgsBadDelim {
             span: dspan.entire(),
             sugg: errors::MacroArgsBadDelimSugg { open: dspan.open, close: dspan.close },
+            rule_kw,
         });
     }
 }
 
+fn check_args_empty(sess: &Session, args: &tokenstream::TokenTree) -> Result<(), ErrorGuaranteed> {
+    match args {
+        tokenstream::TokenTree::Delimited(.., delimited) if delimited.is_empty() => Ok(()),
+        _ => {
+            let msg = "`derive` rules do not accept arguments; `derive` must be followed by `()`";
+            Err(sess.dcx().span_err(args.span(), msg))
+        }
+    }
+}
+
 fn check_lhs(sess: &Session, node_id: NodeId, lhs: &mbe::TokenTree) -> Result<(), ErrorGuaranteed> {
     let e1 = check_lhs_nt_follows(sess, node_id, lhs);
     let e2 = check_lhs_no_empty_seq(sess, slice::from_ref(lhs));