fix: resolve critical linking failures in decomposed modules

Fixed symbol visibility issues preventing proper module linking:

1. Fixed mlir_backend.f90 re-export module to use explicit imports
   - Changed from generic 'use module' to 'use module, only: symbols'
   - Ensures all symbols are properly imported and re-exported
   - Resolves "undefined reference to generate_mlir_module" error

2. Fixed fortran_backend.f90 re-export module similarly
   - Applied same explicit import pattern for consistency
   - Prevents future linking issues

3. Fixed format string syntax error in hlfir_dialect.f90
   - Corrected write statement format specifier
   - Removed extra 'A' from '(A,L1,A,L1,A,L1,A)' -> '(A,L1,A,L1,A,L1)'

4. Fixed performance_benchmarks.f90 test
   - Replaced non-existent destroy_memory_tracker() call
   - Changed to tracker%cleanup() method call

All changes maintain API compatibility while ensuring proper symbol export.

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: krystophny

src/backend/fortran/fortran_backend.f90

@@ -1,7 +1,17 @@
 module fortran_backend
-    use fortran_backend_core
-    use fortran_backend_statements
-    use fortran_backend_expressions
+    use fortran_backend_core, only: fortran_backend_t, generate_code_from_arena, generate_code_program, &
+                                    generate_code_function_def, generate_code_subroutine_def
+    use fortran_backend_statements, only: generate_code_assignment, generate_code_print_statement, &
+                                          generate_code_declaration, generate_code_parameter_declaration, &
+                                          generate_code_if, generate_code_do_loop, generate_code_do_while, &
+                                          generate_code_select_case, generate_code_stop, generate_code_return, &
+                                          generate_code_cycle, generate_code_exit, generate_code_where, &
+                                          generate_code_use_statement, generate_code_subroutine_call, &
+                                          generate_grouped_body, can_group_declarations, can_group_parameters, &
+                                          generate_grouped_declaration, build_param_name_with_dims
+    use fortran_backend_expressions, only: generate_code_literal, generate_code_identifier, &
+                                           generate_code_binary_op, generate_code_call_or_subscript, &
+                                           generate_code_array_literal, find_node_index_in_arena, same_node
     implicit none
     private
 

src/backend/mlir/mlir_backend.f90

@@ -1,10 +1,29 @@
 module mlir_backend
-    use mlir_backend_core
-    use mlir_backend_helpers
-    use mlir_backend_functions
-    use mlir_backend_statements  
-    use mlir_backend_operators
-    use mlir_backend_output
+    use mlir_backend_core, only: generate_mlir_program, generate_mlir_node, generate_mlir_module, &
+                                 generate_mlir_interface_block, generate_mlir_function_declaration, &
+                                 generate_mlir_subroutine_declaration, generate_mlir_module_node, &
+                                 generate_mlir_use_statement, resolve_module_symbols, &
+                                 generate_mlir_where_construct, generate_mlir_derived_type
+    use mlir_backend_helpers, only: format_string_to_array, generate_hlfir_constant, generate_hlfir_load, &
+                                    generate_hlfir_string_literal, int_to_char
+    use mlir_backend_functions, only: generate_mlir_program_functions, generate_mlir_function, &
+                                      generate_mlir_subroutine, generate_function_parameter_list, &
+                                      generate_function_call_with_args, is_function_name
+    use mlir_backend_statements, only: generate_mlir_declaration, generate_mlir_assignment, &
+                                       generate_mlir_pointer_assignment, is_pointer_variable, &
+                                       generate_mlir_subroutine_call, generate_mlir_return, &
+                                       generate_mlir_exit, generate_mlir_cycle, &
+                                       generate_mlir_allocate_statement, generate_array_size_calculation, &
+                                       generate_mlir_deallocate_statement
+    use mlir_backend_operators, only: is_generic_procedure_call, generate_mlir_generic_procedure_call, &
+                                      find_interface_block, resolve_generic_procedure, &
+                                      is_operator_overloaded, generate_mlir_operator_overload_call, &
+                                      generate_mlir_assignment_overload_call, resolve_operator_procedure, &
+                                      resolve_assignment_procedure, get_procedure_name_from_node
+    use mlir_backend_output, only: generate_string_runtime_output, generate_integer_runtime_output, &
+                                   generate_real_runtime_output, generate_logical_runtime_output, &
+                                   generate_variable_runtime_output, generate_expression_runtime_output, &
+                                   generate_call_runtime_output
     implicit none
     private
 

src/dialects/hlfir_dialect.f90

@@ -194,7 +194,7 @@ function create_hlfir_fortran_attrs(context, contiguous, target, optional) resul
         character(len=64) :: attr_str
 
         ! Create a string representation of attributes
-        write(attr_str, '(A,L1,A,L1,A,L1,A)') &
+        write(attr_str, '(A,L1,A,L1,A,L1)') &
             'contiguous=', contiguous, &
             ',target=', target, &
             ',optional=', optional

test/performance_benchmarks.f90

@@ -277,7 +277,7 @@ function benchmark_memory_patterns() result(passed)
         print '(A,F8.3,A)', "  Rate: ", real(iterations)/elapsed, " operations/second"
         print '(A,I0)', "  Peak tracked allocations: ", tracker%get_peak_usage()
 
-        call destroy_memory_tracker(tracker)
+        call tracker%cleanup()
 
         passed = elapsed < 10.0  ! Should complete in under 10 seconds
     end function benchmark_memory_patterns