compiler: Fix some vestigial +0.0/-0.0 issues

[jhogberg]

Fixes #7901

[github-actions]

CT Test Results

       2 files     323 suites   8m 48s ⏱️
   780 tests    778 ✔️ 2 💤 0 ❌
5 326 runs  5 324 ✔️ 2 💤 0 ❌

Results for commit 34ef50c.

♻️ This comment has been updated with latest results.

To speed up review, make sure that you have read Contributing to Erlang/OTP and that all checks pass.

See the TESTING and DEVELOPMENT HowTo guides for details about how to run test locally.

Artifacts

• Complete CT logs (Download Logs)
• HTML Documentation (Download HTML Docs)
• Windows Installer

// Erlang/OTP Github Action Bot

[RobinMorisset]

Thanks for this quick fix.
Some variants of the original bug remain, e.g.:

-export([f2/0, start/0]).

f2() ->
    [ok || _ := _ <- ok, f3(catch f3(ok, -1), -0.0)].

f3(_, 18446744073709551615) ->
    ok.

start() ->
    f3(ok, ok).

Still crashes with

test98296:1: function '-f2/0-lc$^0/1-0-'/1+17:
  Internal consistency check failed - please report this bug.
  Instruction: {call_last,2,{f,9},1}
  Error:       {bad_arg_type,{x,1},
                             {t_float,{-0.0,-0.0}},
                             {t_union,{t_atom,[ok]},
                                      none,
                                      {t_number,{-1,0}},
                                      none,none}}:

[jhogberg]

Thanks, I'll get right on it :)

[RobinMorisset]

This PR also seems to have introduced a new bug:

-export([f4/1]).

f4(_V1) ->
    -0.0 = abs(_V1).

Crashes erlc with

Function: f4/1
Sub pass ssa_opt_type_start
/home/rmorisset/minimized/test98057.erl: internal error in pass beam_ssa_opt:
exception error: no match of right hand side value false
  in function  beam_types:glb_ranges/2 (beam_types.erl, line 933)
  in call from beam_types:glb/2 (beam_types.erl, line 879)
  in call from beam_ssa_type:meet_types/2 (beam_ssa_type.erl, line 2896)
  in call from beam_ssa_type:infer_types_br_1/3 (beam_ssa_type.erl, line 2512)
  in call from beam_ssa_type:infer_types_br/4 (beam_ssa_type.erl, line 2477)
  in call from beam_ssa_type:update_successors/5 (beam_ssa_type.erl, line 1968)
  in call from beam_ssa_type:sig_bs/8 (beam_ssa_type.erl, line 258)
  in call from beam_ssa_type:sig_function_1/4 (beam_ssa_type.erl, line 221)

but only with this PR.

[RobinMorisset]

This PR also seems to have introduced a new bug:

Two other testcases with the same symptom:

-export([f7/1]).

f7(_V1) when -0.0 < floor(_V1 band 1) ->
    ok.

and

-export([f9/0]).

f9() ->
    -0.0 =
        case
            fun() ->
                ok
            end
        of
            _V2 when (_V2 > ok) ->
                2147483647.0;
            _ ->
                -2147483648
        end * 0.

[jhogberg]

Thanks, I had introduced some inconsistencies in the lattice, I've pushed up a new version now :)

[RobinMorisset]

erlfuzz has run for more than 1h with no bug found (it was finding one every 5 to 10mn before this latest fix), so it seems to be fixed for good, thanks!