fix: keep static globals discoverable

- handle DW_TAG_variable entries nested under subprograms
  when DW_AT_external marks them as statics
- add demangled aliases for globals and deduplicate suffix matches
- expand the Rust fixture/test to ensure DWARF can resolve GLOBAL_ENUM

Author: swananan

ghostscope-dwarf/src/core/demangle.rs

@@ -5,23 +5,32 @@ use gimli::DwLang;
 /// Demangle a symbol string using language hint when available.
 /// Returns None if demangling fails or is not applicable.
 pub fn demangle_by_lang(lang: Option<DwLang>, s: &str) -> Option<String> {
+    // 1) Trust DW_AT_language when available
     match lang {
-        Some(gimli::DW_LANG_Rust) => demangle_rust(s),
+        Some(gimli::DW_LANG_Rust) => {
+            if let Some(d) = demangle_rust(s) {
+                return Some(d);
+            }
+        }
         Some(gimli::DW_LANG_C_plus_plus)
         | Some(gimli::DW_LANG_C_plus_plus_11)
         | Some(gimli::DW_LANG_C_plus_plus_14)
         | Some(gimli::DW_LANG_C_plus_plus_17)
-        | Some(gimli::DW_LANG_C_plus_plus_20) => demangle_cpp(s),
-        _ => {
-            // Try common patterns heuristically
-            if is_rust_mangled(s) {
-                demangle_rust(s)
-            } else if is_itanium_cpp_mangled(s) {
-                demangle_cpp(s)
-            } else {
-                None
+        | Some(gimli::DW_LANG_C_plus_plus_20) => {
+            if let Some(d) = demangle_cpp(s) {
+                return Some(d);
             }
         }
+        _ => {}
+    }
+
+    // 2) Fall back to heuristics if language hint missing or demangle failed
+    if is_rust_mangled(s) || looks_like_legacy_rust(s) {
+        demangle_rust(s)
+    } else if is_itanium_cpp_mangled(s) {
+        demangle_cpp(s)
+    } else {
+        None
     }
 }
 
@@ -53,7 +62,11 @@ pub fn demangled_leaf(full: &str) -> String {
 
 /// Heuristic: Rust v0 mangling starts with "_R".
 pub fn is_rust_mangled(s: &str) -> bool {
-    s.starts_with("_R")
+    s.starts_with("_R") || looks_like_legacy_rust(s)
+}
+
+fn looks_like_legacy_rust(s: &str) -> bool {
+    s.starts_with("_ZN") && s.contains("17h") && s.ends_with('E')
 }
 
 /// Heuristic: Itanium C++ mangling starts with "_Z".

ghostscope-dwarf/src/data/lightweight_index.rs

@@ -6,7 +6,7 @@
 //! - Support for parallel construction with index shards
 //! - Fast binary search for symbol lookup
 
-use crate::core::IndexEntry;
+use crate::core::{demangle_by_lang, demangled_leaf, IndexEntry};
 use gimli::{DebugInfoOffset, EndianArcSlice, LittleEndian};
 use std::collections::{BTreeMap, HashMap};
 use tracing::debug;
@@ -89,11 +89,11 @@ impl LightweightIndex {
         }
 
         // Add all type entries (struct/class/union/enum)
-        for (name, ty_entries) in types {
+        for (name, ty_entries) in &types {
             let start_idx = entries.len();
-            entries.extend(ty_entries);
+            entries.extend(ty_entries.clone());
             let indices: Vec<usize> = (start_idx..entries.len()).collect();
-            type_map.insert(name, indices);
+            type_map.insert(name.clone(), indices);
         }
 
         // IMPORTANT: Do NOT sort entries! This would invalidate the indices
@@ -118,6 +118,25 @@ impl LightweightIndex {
             total_functions, total_variables, entries.len(), address_map.len()
         );
 
+        // Ensure demangled aliases exist for variables even if DW_AT_name was missing.
+        for (idx, entry) in entries.iter().enumerate() {
+            if entry.tag == gimli::constants::DW_TAG_variable {
+                if let Some(demangled) = demangle_by_lang(entry.language, entry.name.as_ref()) {
+                    let leaf = demangled_leaf(&demangled);
+                    if leaf != entry.name.as_ref() {
+                        tracing::trace!(
+                            "LightweightIndex: alias '{}' -> '{}' (idx {}, lang={:?})",
+                            entry.name,
+                            leaf,
+                            idx,
+                            entry.language
+                        );
+                        variable_map.entry(leaf).or_default().push(idx);
+                    }
+                }
+            }
+        }
+
         Self {
             entries,
             function_map,

ghostscope-dwarf/src/module/data.rs

@@ -2155,10 +2155,17 @@ impl ModuleData {
         };
 
         let mut out = Vec::new();
+        // Track DIEs we've already emitted (unit_offset, die_offset)
+        let mut seen_offsets: HashSet<(u64, u64)> = HashSet::new();
+
         // Try demangled full (preserve the demangled name that matched)
         if let Some(indices) = self.demangled_variable_map.get(name) {
             for &idx in indices {
                 if let Some(entry) = self.lightweight_index.entry(idx) {
+                    let key = (entry.unit_offset.0 as u64, entry.die_offset.0 as u64);
+                    if !seen_offsets.insert(key) {
+                        continue;
+                    }
                     let link_address = entry.address_ranges.first().and_then(|(lo, hi)| {
                         if lo == hi {
                             Some(*lo)
@@ -2185,6 +2192,10 @@ impl ModuleData {
         if let Some(indices) = self.demangled_variable_leaf_map.get(name) {
             for &idx in indices {
                 if let Some(entry) = self.lightweight_index.entry(idx) {
+                    let key = (entry.unit_offset.0 as u64, entry.die_offset.0 as u64);
+                    if !seen_offsets.insert(key) {
+                        continue;
+                    }
                     let link_address = entry.address_ranges.first().and_then(|(lo, hi)| {
                         if lo == hi {
                             Some(*lo)
@@ -2213,6 +2224,10 @@ impl ModuleData {
         for key in self.lightweight_index.get_variable_names() {
             if key.rsplit("::").next().map(|s| s == name).unwrap_or(false) {
                 for e in self.lightweight_index.find_variables_by_name(key) {
+                    let key = (e.unit_offset.0 as u64, e.die_offset.0 as u64);
+                    if !seen_offsets.insert(key) {
+                        continue;
+                    }
                     let link_address =
                         e.address_ranges
                             .first()
@@ -2242,6 +2257,10 @@ impl ModuleData {
                 if e.tag != gimli::constants::DW_TAG_variable {
                     continue;
                 }
+                let key_offsets = (e.unit_offset.0 as u64, e.die_offset.0 as u64);
+                if !seen_offsets.insert(key_offsets) {
+                    continue;
+                }
                 let last = e.name.rsplit("::").next().unwrap_or(e.name.as_ref());
                 if last == name || e.name == name.into() {
                     let link_address =
@@ -2378,13 +2397,18 @@ impl ModuleData {
     pub(crate) fn find_global_variables_by_name(&self, name: &str) -> Vec<GlobalVariableInfo> {
         let mut out = Vec::new();
         let entries = self.lightweight_index.find_variables_by_name(name);
+        let mut seen_offsets: HashSet<(u64, u64)> = HashSet::new();
 
         // Parse object file once for section classification
         let obj = match object::File::parse(&self._binary_mapped_file.data[..]) {
             Ok(f) => f,
             Err(_) => {
                 // Cannot classify sections, but still return entries with link_address
                 for e in entries {
+                    let key = (e.unit_offset.0 as u64, e.die_offset.0 as u64);
+                    if !seen_offsets.insert(key) {
+                        continue;
+                    }
                     let link_address =
                         e.address_ranges
                             .first()
@@ -2402,6 +2426,10 @@ impl ModuleData {
         };
 
         for e in entries {
+            let key = (e.unit_offset.0 as u64, e.die_offset.0 as u64);
+            if !seen_offsets.insert(key) {
+                continue;
+            }
             let link_address =
                 e.address_ranges
                     .first()

ghostscope-dwarf/src/parser/fast_parser.rs

@@ -1,7 +1,10 @@
 //! Unified DWARF parser - true single-pass parsing
 
 use crate::{
-    core::{IndexEntry, Result},
+    core::{
+        demangle::{demangle_by_lang, demangled_leaf},
+        IndexEntry, Result,
+    },
     data::{
         directory_from_index, resolve_file_path, LightweightFileIndex, LightweightIndex,
         LineMappingTable, ScopedFileIndexManager,
@@ -246,65 +249,118 @@ impl<'a> DwarfParser<'a> {
                     }
                 }
                 gimli::constants::DW_TAG_variable => {
+                    tracing::trace!(
+                        "Evaluating global variable DIE {:?} in CU {:?}",
+                        entry.offset(),
+                        unit_offset
+                    );
+                    let is_static_symbol = self.is_static_symbol(entry).unwrap_or(false);
                     let in_function_scope = tag_stack.iter().any(|t| {
                         *t == gimli::constants::DW_TAG_subprogram
                             || *t == gimli::constants::DW_TAG_inlined_subroutine
                     });
-                    if in_function_scope {
+                    if in_function_scope && !is_static_symbol {
+                        tracing::trace!(
+                            "Skipping variable at {:?} (in function scope, stack={:?})",
+                            entry.offset(),
+                            tag_stack
+                        );
                         // Skip local variables
                         tag_stack.push(entry.tag());
                         continue;
+                    } else if in_function_scope {
+                        // Rust (and some C compilers) sometimes nest file-scoped statics under the
+                        // function that first references them, even though DW_AT_location uses
+                        // DW_OP_addr. When DW_AT_external is false we treat them as true globals.
+                        tracing::trace!(
+                            "Treating static variable at {:?} as global despite function scope (stack={:?})",
+                            entry.offset(),
+                            tag_stack
+                        );
                     }
                     if Self::is_declaration(entry).unwrap_or(false) {
+                        tracing::trace!(
+                            "Skipping variable at {:?} (declaration-only DIE)",
+                            entry.offset()
+                        );
                         tag_stack.push(entry.tag());
                         continue;
                     }
+                    let mut collected_names: Vec<(String, bool)> = Vec::new();
+                    let mut push_unique_name = |candidate: String, is_linkage_alias: bool| {
+                        if candidate.is_empty() {
+                            return;
+                        }
+                        if collected_names
+                            .iter()
+                            .any(|(existing, _)| existing == &candidate)
+                        {
+                            return;
+                        }
+                        collected_names.push((candidate, is_linkage_alias));
+                    };
+
+                    let mut have_primary_name = false;
                     if let Some(name) = self.extract_name(self.dwarf, unit, entry)? {
-                        let flags = crate::core::IndexFlags {
-                            is_static: self.is_static_symbol(entry).unwrap_or(false),
-                            ..Default::default()
-                        };
-                        // Restore variable address for globals/statics via DW_AT_location
-                        let var_addr = self.extract_variable_address(entry, unit)?;
-                        let var_ranges = var_addr.map(|a| vec![(a, a)]).unwrap_or_default();
+                        push_unique_name(name, false);
+                        have_primary_name = true;
+                    }
+
+                    if let Some((linkage_name, _)) =
+                        self.extract_linkage_name(self.dwarf, unit, entry)?
+                    {
+                        if let Some(demangled) =
+                            demangle_by_lang(cu_language, linkage_name.as_str())
+                        {
+                            let leaf = demangled_leaf(&demangled);
+                            push_unique_name(leaf, false);
+                            have_primary_name = true;
+                        }
+                        push_unique_name(linkage_name.clone(), true);
+                    }
+
+                    if !have_primary_name {
+                        tracing::trace!(
+                            "DWARF variable at {:?} missing usable name (CU lang={:?}); skipping alias registration",
+                            entry.offset(),
+                            cu_language
+                        );
+                        tag_stack.push(entry.tag());
+                        continue;
+                    }
+
+                    let flags = crate::core::IndexFlags {
+                        is_static: is_static_symbol,
+                        ..Default::default()
+                    };
+                    let var_addr = self.extract_variable_address(entry, unit)?;
+                    let var_ranges = var_addr.map(|a| vec![(a, a)]).unwrap_or_default();
+
+                    for (name, is_linkage_alias) in collected_names {
+                        let mut entry_flags = flags;
+                        entry_flags.is_linkage = is_linkage_alias;
                         let index_entry = IndexEntry {
                             name: std::sync::Arc::from(name.as_str()),
                             die_offset: entry.offset(),
                             unit_offset,
                             tag: entry.tag(),
-                            flags,
+                            flags: entry_flags,
                             language: cu_language,
                             address_ranges: var_ranges.clone(),
                             entry_pc: None,
                         };
+                        tracing::trace!(
+                            "Registering variable alias '{}' (linkage={}, lang={:?}, die={:?})",
+                            name,
+                            entry_flags.is_linkage,
+                            cu_language,
+                            entry.offset()
+                        );
                         shard
                             .variables
                             .entry(name.clone())
                             .or_default()
                             .push(index_entry);
-                        if let Some((linkage_name, _)) =
-                            self.extract_linkage_name(self.dwarf, unit, entry)?
-                        {
-                            if linkage_name != name {
-                                let mut alias_flags = flags;
-                                alias_flags.is_linkage = true;
-                                let index_entry_linkage = IndexEntry {
-                                    name: std::sync::Arc::from(linkage_name.as_str()),
-                                    die_offset: entry.offset(),
-                                    unit_offset,
-                                    tag: entry.tag(),
-                                    flags: alias_flags,
-                                    language: cu_language,
-                                    address_ranges: var_ranges,
-                                    entry_pc: None,
-                                };
-                                shard
-                                    .variables
-                                    .entry(linkage_name)
-                                    .or_default()
-                                    .push(index_entry_linkage);
-                            }
-                        }
                     }
                 }
                 gimli::constants::DW_TAG_structure_type