canonicalize names of nested functions and nested lambdas

Author: d-netto

doc/src/manual/functions.md

@@ -292,12 +292,12 @@ syntaxes:
 
 ```jldoctest
 julia> x -> x^2 + 2x - 1
-#1 (generic function with 1 method)
+#2 (generic function with 1 method)
 
 julia> function (x)
            x^2 + 2x - 1
        end
-#3 (generic function with 1 method)
+#5 (generic function with 1 method)
 ```
 
 Each statement creates a function taking one argument `x` and returning the value of the polynomial `x^2 +

src/ast.c

@@ -7,6 +7,9 @@
 #include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
+
+#include "support/strhash.h"
+
 #ifdef _OS_WINDOWS_
 #include <malloc.h>
 #endif
@@ -215,11 +218,85 @@ static value_t fl_nothrow_julia_global(fl_context_t *fl_ctx, value_t *args, uint
         decode_restriction_value(pku) : jl_atomic_load_relaxed(&b->value)) != NULL ? fl_ctx->T : fl_ctx->F;
 }
 
+static htable_t *rebuild_counter_table(jl_module_t *m) JL_NOTSAFEPOINT
+{
+    htable_t *counter_table = m->counter_table = htable_new((htable_t *)malloc_s(sizeof(htable_t)), 0);
+    jl_svec_t *t = jl_atomic_load_relaxed(&m->bindings);
+    for (size_t i = 0; i < jl_svec_len(t); i++) {
+        jl_binding_t *b = (jl_binding_t*)jl_svecref(t, i);
+        if ((void*)b == jl_nothing) {
+            continue;
+        }
+        char *globalref_name = jl_symbol_name(b->globalref->name);
+        if (is_canonicalized_anonfn_typename(globalref_name)) {
+            int should_free = 1;
+            // copy globalref_name into the buffer until we hit a `#` character
+            char *delim = strchr(&globalref_name[1], '#');
+            assert(delim != NULL);
+            size_t len = delim - globalref_name - 1;
+            assert(len > 0);
+            char *enclosing_function_name = (char*)calloc_s(len + 1);
+            memcpy(enclosing_function_name, &globalref_name[1], len);
+            // check if the enclosing function name is already in the counter table
+            if (strhash_get(counter_table, enclosing_function_name) == HT_NOTFOUND) {
+                strhash_put(counter_table, enclosing_function_name, (void*)((uintptr_t)HT_NOTFOUND + 1));
+                should_free = 0;
+            }
+            char *pint = strrchr(globalref_name, '#');
+            assert(pint != NULL);
+            int counter = atoi(pint + 1);
+            int max_seen_so_far = ((uint32_t)(uintptr_t)strhash_get(counter_table, enclosing_function_name) - (uintptr_t)HT_NOTFOUND - 1);
+            if (counter >= max_seen_so_far) {
+                strhash_put(counter_table, enclosing_function_name, (void*)((uintptr_t)counter + 1 + (uintptr_t)HT_NOTFOUND + 1));
+            }
+            if (should_free) {
+                free(enclosing_function_name);
+            }
+        }
+    }
+    return counter_table;
+}
+
+// used to generate a unique suffix for a given symbol (e.g. variable or type name)
+// first argument contains a stack of method definitions seen so far by `closure-convert` in flisp.
+// if the top of the stack is non-NIL, we use it to augment the suffix so that it becomes
+// of the form $top_level_method_name##$counter, where counter is stored in a per-module
+// side table indexed by top-level method name.
+// this ensures that precompile statements are a bit more stable across different versions
+// of a codebase. see #53719
 static value_t fl_current_module_counter(fl_context_t *fl_ctx, value_t *args, uint32_t nargs) JL_NOTSAFEPOINT
 {
+    argcount(fl_ctx, "current-julia-module-counter", nargs, 1);
     jl_ast_context_t *ctx = jl_ast_ctx(fl_ctx);
-    assert(ctx->module);
-    return fixnum(jl_module_next_counter(ctx->module));
+    jl_module_t *m = ctx->module;
+    assert(m != NULL);
+    // Get the outermost function name from the `parsed_method_stack` top
+    char *funcname = NULL;
+    value_t parsed_method_stack = args[0];
+    if (parsed_method_stack != fl_ctx->NIL) {
+        value_t bottom_stack_symbol = fl_applyn(fl_ctx, 1, symbol_value(symbol(fl_ctx, "last")), parsed_method_stack);
+        funcname = symbol_name(fl_ctx, bottom_stack_symbol);
+    }
+    char buf[(funcname != NULL ? strlen(funcname) : 0) + 20];
+    if (funcname != NULL && strchr(funcname, '#') == NULL) {
+        jl_mutex_lock_nogc(&m->lock);
+        htable_t *counter_table = m->counter_table;
+        if (counter_table == NULL) {
+            counter_table = rebuild_counter_table(m);
+        }
+        // try to find the function name in the module's counter table, if it's not found, add it
+        if (strhash_get(counter_table, funcname) == HT_NOTFOUND) {
+            strhash_put(counter_table, funcname, (void*)((uintptr_t)HT_NOTFOUND + 1));
+        }
+        uint32_t nxt = ((uint32_t)(uintptr_t)strhash_get(counter_table, funcname) - (uintptr_t)HT_NOTFOUND - 1);
+        snprintf(buf, sizeof(buf), "%s##%d", funcname, nxt);
+        strhash_put(counter_table, funcname, (void*)(nxt + 1 + (uintptr_t)HT_NOTFOUND + 1));
+        jl_mutex_unlock_nogc(&m->lock);
+    }
+    else {
+        snprintf(buf, sizeof(buf), "%d", jl_module_next_counter(ctx->module));
+    }
+    return symbol(fl_ctx, buf);
 }
 
 static int jl_is_number(jl_value_t *v)

src/datatype.c

@@ -20,23 +20,21 @@ extern "C" {
 
 // allocating TypeNames -----------------------------------------------------------
 
-static int is10digit(char c) JL_NOTSAFEPOINT
-{
-    return (c >= '0' && c <= '9');
-}
-
 static jl_sym_t *jl_demangle_typename(jl_sym_t *s) JL_NOTSAFEPOINT
 {
     char *n = jl_symbol_name(s);
     if (n[0] != '#')
         return s;
-    char *end = strrchr(n, '#');
+    char *end = strchr(&n[1], '#');
+    // handle `#f...##...#...`
+    if (end != NULL && end[1] == '#')
+        end = strchr(&end[2], '#');
     int32_t len;
-    if (end == n || end == n+1)
+    if (end == NULL || end == n+1)
         len = strlen(n) - 1;
     else
         len = (end-n) - 1;  // extract `f` from `#f#...`
-    if (is10digit(n[1]))
+    if (isdigit(n[1]) || is_canonicalized_anonfn_typename(n))
         return _jl_symbol(n, len+1);
     return _jl_symbol(&n[1], len);
 }

src/flisp/flisp.h

@@ -158,7 +158,7 @@ value_t fl_cons(fl_context_t *fl_ctx, value_t a, value_t b) JL_NOTSAFEPOINT;
 value_t fl_list2(fl_context_t *fl_ctx, value_t a, value_t b) JL_NOTSAFEPOINT;
 value_t fl_listn(fl_context_t *fl_ctx, size_t n, ...) JL_NOTSAFEPOINT;
 value_t symbol(fl_context_t *fl_ctx, const char *str) JL_NOTSAFEPOINT;
-char *symbol_name(fl_context_t *fl_ctx, value_t v);
+char *symbol_name(fl_context_t *fl_ctx, value_t v) JL_NOTSAFEPOINT;
 int fl_is_keyword_name(const char *str, size_t len);
 value_t alloc_vector(fl_context_t *fl_ctx, size_t n, int init);
 size_t llength(value_t v);

src/init.c

@@ -869,6 +869,7 @@ static NOINLINE void _finish_julia_init(JL_IMAGE_SEARCH rel, jl_ptls_t ptls, jl_
         jl_load(jl_core_module, "boot.jl");
         post_boot_hooks();
     }
+    jl_main_module->counter_table = NULL;
 
     if (jl_base_module == NULL) {
         // nthreads > 1 requires code in Base