Hi,

Thanks for pointing out the bug.

I don't think this is the best correction, though. I'd rather see:

diff --git a/lib/stdlib/src/shell.erl b/lib/stdlib/src/shell.erl
index 6eafc7b..9b72312 100644
--- a/lib/stdlib/src/shell.erl
+++ b/lib/stdlib/src/shell.erl
@@ -758,6 +758,8 @@ used_records({record,,R,Name,Ups}) ->
{name, Name, [R | Ups]};
used_records({record_field,,R,F}) -> % illegal
{expr, [R | F]};
+used_records({typed_record_field,Field,_Type}) ->

• used_records(Field);
  used_records({call,,{atom,,record},[A,{atom,,Name}]}) ->
  {name, Name, A};
  used_records({call,,{atom,,is_record},[A,{atom,,Name}]}) ->

Another case we missed when changing the representation of typed
record fields some time ago.

Best regards

Hi,

[Sorry for the ugly indentation.]

Thanks for pointing out the bug.

I don't think this is the best correction, though. I'd rather see:

diff --git a/lib/stdlib/src/shell.erl b/lib/stdlib/src/shell.erl
index 6eafc7b..9b72312 100644
--- a/lib/stdlib/src/shell.erl
+++ b/lib/stdlib/src/shell.erl
@@ -758,6 +758,8 @@ used_records({record,_,R,Name,Ups}) ->
     {name, Name, [R | Ups]};
 used_records({record_field,_,R,F}) -> % illegal
     {expr, [R | F]};
+used_records({typed_record_field,Field,_Type}) ->
+    used_records(Field);
 used_records({call,_,{atom,_,record},[A,{atom,_,Name}]}) ->
     {name, Name, A};
 used_records({call,_,{atom,_,is_record},[A,{atom,_,Name}]}) ->

Another case we missed when changing the representation of typed
record fields some time ago.

Best regards

@uabboli This certainly looks nicer, but does not work when there are other record fields:

-module(bug).
-export([t/0]).

-record(r0,{f0}).
-record(r,{f :: #r{} | undefined, f0 :: #r0{}}).

t() ->
   #r{f=#r{f=undefined, f0 = #r0{f0 = v}}}.

results in:

Erlang/OTP 20 [RELEASE CANDIDATE 2] [erts-9.0] [source] [64-bit] [smp:8:8] [ds:8:8:10] [async-threads:10] [hipe] [kernel-poll:false]

Eshell V9.0  (abort with ^G)
1> c("bug.erl").
{ok,bug}
2> R = bug:t().
{r,{r,undefined,{r0,v}},undefined}
3> rr(bug).
[r,r0]
4> R#r.f.
* 5: record r0 undefined

with proposed fix result is as expected:

4> R#r.f.
#r{f = undefined,f0 = #r0{f0 = v}}

I see. Forget my last suggestion; it's no good.

Your patch tests if record Name has already been seen. Note however
that you also need to take E1 into consideration: only if E1 has already
been processed, can Name be skipped.

Extending your example a bit:

-module(bug).
-export([t/0, t2/0]).

-record(r0,{f0}).
-record(r,{f :: #r{} | undefined, f0 :: #r0{}}).

t() ->
   #r{f=#r{f=undefined, f0 = #r0{f0 = v}}}.

-record(r2, {}).

t2() ->
    #r2{}.

Now try:

> erl
Eshell V9.0  (abort with ^G)
1> c(bug).
{ok,bug}
2> rr(bug).
[r,r0,r2]
3> (#r{f = #r2{}})#r{f = 18}.
* 1: record r2 undefined
4> 

@uabboli good point
I've updated the pull request to accordingly.
Now:

3> (#r{f = #r2{}})#r{f = 18}.
#r{f = 18,f0 = undefined}

Looks good.

A test in shell_SUITE.erl would be nice. It should be possible to
use, for example, otp_6166/1 as a starting point.

The patch cannot make it into Erlang/OTP 20.0 since the code is
frozen. As soon as the release is out (Wednesday is the plan), you
can rebase the patch on maint.

Added typed_records/1 test in shell_SUITE.erl .

It turns out that if already seen records are always skipped that messes up
the expected order and there are test failures, so I modified the patch to skip
a record only when its own definition is processed.

When Erlang/OTP 20.0 is out I will rebase the patch on maint.

@uabboli rebased onto erlang:maint

Thanks. I'll merge the branch soon.

Merged commit 693691a into erlang:maint.