guix/gnu/packages/patches/intel-xed-fix-nondeterminism.patch

This patch removes sources of build non-determinism in the upstream sources.

In particular, many of the compiled sources are generated with Python code,
which in turn uses dictionaries to index the output C functions.  However,
iterators over Python dictionaries have no guaranteed order, thus resulting in
the C functions being output in a random order between builds.

The patch below fixes this by forcing an order during output in several key
places.  Note, however, that future updates may uncover new such places that
just happen to be non-problematic at the time of this patch.  If you are
reading this due to finding such issues, feel free to contact me at
elaexuotee@wilsonb.com for help.

diff --git a/pysrc/ild_codegen.py b/pysrc/ild_codegen.py
index 628ec45..a9bff79 100755
--- a/pysrc/ild_codegen.py
+++ b/pysrc/ild_codegen.py
@@ -188,14 +188,14 @@ def gen_l2_func_list(agi, target_nt_dict, arg_nt_dict,
                      ild_t_member):
     """generate L2 functions"""
     l2_func_list = []
-    for (nt_name,array) in target_nt_dict.items():
+    for (nt_name,array) in sorted(target_nt_dict.items()):
         target_opname = array.get_target_opname()
         if array.is_const_lookup_fun():
             fo = gen_const_l2_function(agi, nt_name,
                                 target_opname, ild_t_member)
             l2_func_list.append(fo)
         else:
-            for arg_nt_seq,arg_arr in arg_nt_dict.items():
+            for arg_nt_seq,arg_arr in sorted(arg_nt_dict.items()):
                 fo = gen_scalable_l2_function(agi, nt_name,
                      target_opname, ild_t_member, arg_arr, list(arg_nt_seq))
                 l2_func_list.append(fo)
diff --git a/pysrc/ild_disp.py b/pysrc/ild_disp.py
index 942c036..cf80e29 100755
--- a/pysrc/ild_disp.py
+++ b/pysrc/ild_disp.py
@@ -350,7 +350,8 @@ def work(agi, united_lookup,  disp_nts, brdisp_nts, ild_gendir,
     disp_dict = _gen_l3_array_dict(agi, disp_nts, _disp_token)
 
     
-    nt_arr_list = list(brdisp_dict.values()) + list(disp_dict.values())
+    nt_arr_list = ([v for (k,v) in sorted(brdisp_dict.items())] +
+                   [v for (k,v) in sorted(disp_dict.items())])
     #create function that calls all initialization functions
     init_f = ild_nt.gen_init_function(nt_arr_list, 'xed_ild_disp_l3_init')
     
@@ -367,7 +368,7 @@ def work(agi, united_lookup,  disp_nts, brdisp_nts, ild_gendir,
     l2_functions = []
     eosz_op = ild_eosz.get_target_opname()
     easz_op = ild_easz.get_target_opname()
-    for nt_name,array in list(disp_dict.items()) + list(brdisp_dict.items()):
+    for nt_name,array in sorted(disp_dict.items()) + sorted(brdisp_dict.items()):
         #Some DISP NTs depend on EOSZ, others on EASZ, we need to know
         #that when we generate L2 functions
         if eosz_op in array.get_arg_names():
diff --git a/pysrc/ild_easz.py b/pysrc/ild_easz.py
index 02cd691..c53b9f2 100755
--- a/pysrc/ild_easz.py
+++ b/pysrc/ild_easz.py
@@ -165,9 +165,10 @@ def work(agi, united_lookup, easz_nts, ild_gendir, debug):
             return
         nt_seq_arrays[tuple(nt_seq)] = array
     #init function calls all single init functions for the created tables
-    init_f = ild_nt.gen_init_function(list(nt_seq_arrays.values()),
+    nt_seq_values = [v for (k,v) in sorted(nt_seq_arrays.items())]
+    init_f = ild_nt.gen_init_function(nt_seq_values,
                                        'xed_ild_easz_init')
-    ild_nt.dump_lu_arrays(agi, list(nt_seq_arrays.values()), _easz_c_fn, 
+    ild_nt.dump_lu_arrays(agi, nt_seq_values, _easz_c_fn, 
                           mbuild.join('include-private', _easz_header_fn),
                           init_f)
     getter_fos = []
diff --git a/pysrc/ild_eosz.py b/pysrc/ild_eosz.py
index 6643bc3..89d2d89 100755
--- a/pysrc/ild_eosz.py
+++ b/pysrc/ild_eosz.py
@@ -200,10 +200,11 @@ def work(agi, united_lookup, eosz_nts, ild_gendir, debug):
             return None
         nt_seq_arrays[tuple(nt_seq)] = array
     #init function calls all single init functions for the created tables
-    init_f = ild_nt.gen_init_function(list(nt_seq_arrays.values()), 
+    nt_seq_values = [v for (k,v) in sorted(nt_seq_arrays.items())]
+    init_f = ild_nt.gen_init_function(nt_seq_values, 
                                       'xed_ild_eosz_init')
     #dump init and lookup functions for EOSZ sequences
-    ild_nt.dump_lu_arrays(agi, list(nt_seq_arrays.values()), _eosz_c_fn,
+    ild_nt.dump_lu_arrays(agi, nt_seq_values, _eosz_c_fn,
                           mbuild.join('include-private', _eosz_header_fn),
                           init_f)
     #generate EOSZ getter functions - they get xed_decoded_inst_t*
diff --git a/pysrc/ild_imm.py b/pysrc/ild_imm.py
index 51c413c..0530bae 100755
--- a/pysrc/ild_imm.py
+++ b/pysrc/ild_imm.py
@@ -322,12 +322,14 @@ def work(agi, united_lookup, imm_nts, ild_gendir, eosz_dict,
                                      level='l3')
         nt_dict[nt_name] = array
 
+    nt_dict_values = [v for (k,v) in sorted(nt_dict.items())]
+
     #create function that calls all initialization functions for L3
-    init_f = ild_nt.gen_init_function(list(nt_dict.values()),
+    init_f = ild_nt.gen_init_function(nt_dict_values,
                                       'xed_ild_imm_l3_init')
     
     #dump L3 functions
-    ild_nt.dump_lu_arrays(agi, list(nt_dict.values()), _l3_c_fn,
+    ild_nt.dump_lu_arrays(agi, nt_dict_values, _l3_c_fn,
                           mbuild.join('include-private',_l3_header_fn),
                           init_f)