Support OTP 28 #5507

janl · 2025-04-18T10:29:55Z

based on -rc3

janl · 2025-04-18T11:12:47Z

Update: fixed.

My local testing box (macOS 16.4.1/latest, arm64) with OTP-28-rc3 makes this test fail:

module 'chttpd_socket_buffer_size_test'
  chttpd socket_buffer_size_test
    chttpd_socket_buffer_size_test:67: -default_buffer/2-fun-1-...[0.023 s] ok
    chttpd_socket_buffer_size_test:51: -small_recbuf/2-fun-1-...*failed*
in function chttpd_socket_buffer_size_test:'-small_recbuf/2-fun-1-'/1 (test/eunit/chttpd_socket_buffer_size_test.erl, line 54)
in call from eunit_test:run_testfun/1 (eunit_test.erl, line 83)
in call from eunit_proc:run_test/1 (eunit_proc.erl, line 554)
in call from eunit_proc:with_timeout/3 (eunit_proc.erl, line 379)
in call from eunit_proc:handle_test/2 (eunit_proc.erl, line 537)
in call from eunit_proc:tests_inorder/3 (eunit_proc.erl, line 479)
in call from eunit_proc:with_timeout/3 (eunit_proc.erl, line 369)
in call from eunit_proc:run_group/2 (eunit_proc.erl, line 593)
**error:{assert,[{module,chttpd_socket_buffer_size_test},
         {line,54},
         {expression,"Response =:= 400 orelse Response =:= request_failed"},
         {expected,true},
         {value,false}]}
  output:<<"">>

In the logs, the expected-to-fail request shows a 201:

[notice] 2025-04-18T10:09:29.375779Z nonode@nohost <0.639.0> 8925f5c766 127.0.0.1:60869 127.0.0.1 chttpd_db_socket_buffer_size_test_admin PUT /eunit-test-db-145c735070a4bad74540e819cebb134e/xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 201 ok 17

The relevant test bits are just:

                {"[{recbuf, 1024}]", fun small_recbuf/2},

small_recbuf(_, {_, Db}) ->
    {timeout, 30,
        ?_test(begin
            Id = data(2048),
            Response = put_req(url(Db) ++ "/" ++ Id, "{}"),
            ?assert(Response =:= 400 orelse Response =:= request_failed)
        end)}.

I couldn’t find anything obvious in our or the OTP commits that could cause this.

The test passes with OTP-25 and -27 and -28-rc2.

janl · 2025-04-18T13:27:23Z

Next, Mango:

The PCRE2 upgrade notes state:

Compiled Patterns Should Not be Shared: Sharing the result of re:compile/2 across node boundaries or persisting it has never been officially supported, although it might have accidentally worked between certain previous versions. The internal format produced by re:compile/2 has changed, and will not work on older OTP versions or across nodes. Relying on this undocumented behavior must be changed to compile the regular expression on the node instance where it is intended to be used.

This seems to include mochiglobal:put():

  mango_idx_text: indexable_fields_test...*failed*
in function erl_syntax:abstract/1 (erl_syntax.erl, line 7294)
in call from erl_syntax:abstract_list/1 (erl_syntax.erl, line 7297)
in call from erl_syntax:abstract_list/1 (erl_syntax.erl, line 7297)
in call from erl_syntax:abstract/1 (erl_syntax.erl, line 7280)
in call from mochiglobal:term_to_abstract/3 (src/mochiglobal.erl, line 95)
in call from mochiglobal:forms/2 (src/mochiglobal.erl, line 77)
in call from mochiglobal:compile/2 (src/mochiglobal.erl, line 71)
in call from mochiglobal:put/3 (src/mochiglobal.erl, line 51)
**error:{badarg,#Ref<0.2315338412.3477471233.154666>}
  output:<<"">>

This change to mango_util fixes the issues, but now we are not caching the compiled re:

 cached_re(Name, RE) ->
-    case mochiglobal:get(Name) of
-        undefined ->
-            {ok, MP} = re:compile(RE),
-            ok = mochiglobal:put(Name, MP),
-            MP;
-        MP ->
-            MP
-    end.
+    {ok, MP} = re:compile(RE),
+    MP.

janl · 2025-04-18T13:29:58Z

And then still in Mango, we get this a couple of times, seems we expect a reversal of some sort that is no longer happening. I haven’t had time to dig into this, ~~but there were erl_syntax changes around reversal of lists in rc3, will update here when I know more:~~

The issue is that sets is now backed by maps by default and that changes the order of sets:to_list(), something we should not have relied on in the first place:

sets version 1:

Erlang/OTP 27 [erts-15.2.3] [source] [64-bit] [smp:14:14] [ds:14:14:10] [async-threads:1] [jit]

Eshell V15.2.3 (press Ctrl+G to abort, type help(). for help)
1> A = sets:from_list(["a","b", "c"]).
{set,3,16,16,8,80,48,
     {[],[],[],[],[],[],[],[],[],[],[],[],[],[],[],[]},
     {{[],
       ["b"],
       [],[],[],[],
       ["c"],
       [],[],[],[],[],
       ["a"],
       [],[],[]}}}
2> B = sets:from_list(["b"]).
{set,1,16,16,8,80,48,
     {[],[],[],[],[],[],[],[],[],[],[],[],[],[],[],[]},
     {{[],["b"],[],[],[],[],[],[],[],[],[],[],[],[],[],[]}}}
3> sets:to_list(sets:subtract(A, B)).
["c","a"]

sets version 2:

Erlang/OTP 27 [erts-15.2.3] [source] [64-bit] [smp:14:14] [ds:14:14:10] [async-threads:1] [jit]

Eshell V15.2.3 (press Ctrl+G to abort, type help(). for help)
1> A = sets:from_list(["a","b", "c"], [{version, 2}]).
#{"a" => [],"b" => [],"c" => []}
2> B = sets:from_list(["b"], [{version, 2}]).
#{"b" => []}
3> sets:to_list(sets:subtract(A, B)).
["a","c"]

I pushed a quick fix that helps us pass the tests, but we might have to revisit, if we accidentally rely on the representation elsewhere.

 mango_cursor:1110: -extract_selector_hints_test_/0-fun-4- (t_extract_selector_hints_view)...*failed*
in function mango_cursor:t_extract_selector_hints_view/1 (src/mango_cursor.erl, line 1129)
in call from eunit_test:run_testfun/1 (eunit_test.erl, line 83)
in call from eunit_proc:run_test/1 (eunit_proc.erl, line 554)
in call from eunit_proc:with_timeout/3 (eunit_proc.erl, line 379)
in call from eunit_proc:handle_test/2 (eunit_proc.erl, line 537)
in call from eunit_proc:tests_inorder/3 (eunit_proc.erl, line 479)
in call from eunit_proc:with_timeout/3 (eunit_proc.erl, line 369)
in call from eunit_proc:run_group/2 (eunit_proc.erl, line 593)
**error:{assertEqual,[{module,mango_cursor},
              {line,1129},
              {expression,"extract_selector_hints ( selector )"},
              {expected,[{[{type,json},
                           {indexable_fields,["field2"]},
                           {unindexable_fields,["field3","field1"]}]}]},
              {value,[{[{type,json},
                        {indexable_fields,["field2"]},
                        {unindexable_fields,["field1","field3"]}]}]}]}
  output:<<"">>

nickva · 2025-04-21T02:00:21Z

src/mango/src/mango_cursor.erl

+expect_set_from_list_ordered(L) ->
+    case erlang:system_info(otp_release) of
+        N when N >= "28" -> lists:sort(L);
+        _ -> L
+    end.
+


Instead of hard-coding OTP 28 checks in the test, the test should be improved to check that either the sets match or the the sorted lists of expected vs actual unindexable_fields match.

I gave it a try here #5510. It's a general improvement regardless of OTP version that just make us use all new sets and sorts the results to they are deterministic, otherwise tests might fail randomly again in the future

nickva · 2025-04-22T04:27:08Z

This pr should fix the mochiglobal issue. Thanks to @jaydoane for the review

With OTP 28 sets switchd to using version 2 by default. That caused various tests to fail, as Jan discovered in #5507 In the previous PR #5510 we switched few mango instances to use a helper utility, but it was repeated a few times and Gábor made a good suggestion to use utility functions instead, so that's what this PR does. But it does it for the whole code base. Once we're post OTP 28 we can remove the utility and just use plain set functions.

…dinlgy

janl · 2025-04-24T07:44:21Z

cleaned up the WIP work that @nickva has been properly fixing in the referenced commit. This should be ready to go now.

big-r81

Nice work together! +1

nickva

+1 as well for a good team effort!

I see the elixir error only, same one found by Jan

00:43:38.923 [error] beam/beam_load.c(594): Error loading function 'Elixir.Hex.State':print_invalid_config_error/2: op bs_add p x x u x:
  please re-compile this module with an Erlang/OTP 28 compiler or update your Erlang/OTP version

That seems to be tracked in elixir-lang/elixir#14349

So it seems we'll just need to wait for OTP 28 to be released and the hex issue fixed to be able to add OTP 28 proper to the CI and build chain.

big-r81 approved these changes Apr 18, 2025

View reviewed changes

janl marked this pull request as draft April 18, 2025 11:39

nickva reviewed Apr 21, 2025

View reviewed changes

This was referenced Apr 22, 2025

Use persistent terms to cache Mango RE expressions #5509

Merged

Use new sets only in mango #5510

Merged

nickva mentioned this pull request Apr 23, 2025

Use new sets everywhere #5514

Merged

janl added 3 commits April 24, 2025 09:43

feat: support OTP 28, original patch by @rnewson

1b4f0e6

fix: rebar xref output being reordered

00c5178

fix: Erlang buffer sizes have changed, adjust small recbuf test accor…

bb90182

…dinlgy

janl force-pushed the feat/otp-28 branch from f750900 to bb90182 Compare April 24, 2025 07:43

janl marked this pull request as ready for review April 24, 2025 07:44

big-r81 approved these changes Apr 24, 2025

View reviewed changes

nickva approved these changes Apr 24, 2025

View reviewed changes

janl merged commit 17bf647 into main Apr 24, 2025
23 checks passed

janl deleted the feat/otp-28 branch April 24, 2025 14:21

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Support OTP 28 #5507

Support OTP 28 #5507

janl commented Apr 18, 2025

janl commented Apr 18, 2025 •

edited

Loading

janl commented Apr 18, 2025

janl commented Apr 18, 2025 •

edited

Loading

nickva Apr 21, 2025

nickva Apr 22, 2025

nickva commented Apr 22, 2025

janl commented Apr 24, 2025

big-r81 left a comment

nickva left a comment

Support OTP 28 #5507

Support OTP 28 #5507

Conversation

janl commented Apr 18, 2025

janl commented Apr 18, 2025 • edited Loading

janl commented Apr 18, 2025

janl commented Apr 18, 2025 • edited Loading

nickva Apr 21, 2025

Choose a reason for hiding this comment

nickva Apr 22, 2025

Choose a reason for hiding this comment

nickva commented Apr 22, 2025

janl commented Apr 24, 2025

big-r81 left a comment

Choose a reason for hiding this comment

nickva left a comment

Choose a reason for hiding this comment

janl commented Apr 18, 2025 •

edited

Loading

janl commented Apr 18, 2025 •

edited

Loading