While reviewing PR8222, I found err_ret->text is assigned a not malloced string, but it will finally get freed in err_input.

I expected it should be easy to reproduce a crash, but I failed to find an example.

This issue is related to bpo-34080.

I expected it should be easy to reproduce a crash, but I failed to find an example.

I don't understand how parsetok.c is supposed to handle "from __future__ import barry_as_FLUFL":

• Parser/parsetok.c tests (ps->p_flags & CO_FUTURE_BARRY_AS_BDFL)
• Python/future.c uses "if (strcmp(feature, FUTURE_BARRY_AS_BDFL) == 0) { ff->ff_features |= CO_FUTURE_BARRY_AS_BDFL; }"
• Python/pythonrun.c: PARSER_FLAGS() copies CO_FUTURE_BARRY_AS_BDFL flag as PyPARSE_BARRY_AS_BDFL

The only way to use "<>" operator seems to pass explicitly CO_FUTURE_BARRY_AS_BDFL flag to compile() flags, which is not the convenient way to use "from __future__ import barry_as_FLUFL" :-(

Code:
---

import inspect
CO_FUTURE_BARRY_AS_BDFL = 0x40000
flags = CO_FUTURE_BARRY_AS_BDFL
code = "print(1 <> 2)"
try:
    compile(code, "filename", "exec", flags=0)
except SyntaxError:
    print("ok")
else:
    raise Exception("SyntaxError expected")
compile(code, "filename", "exec", flags=flags)
print("ok")

Maybe we need a test for the easter egg. Or maybe we should remove it? :-p

Oh, this easter egg is tested by Lib/test/test_flufl.py.

Could you add a test in test_flufl.py that will fail with not patched code?

This issue and bpo-34080 look unrelated to me. They can be fixed independently.

Could you add a test in test_flufl.py that will fail with not patched code?

It seems like PyObject_FREE() is never called with the static string "with Barry as BDFL, use '<>' instead of '!='".

When parsetok() goes to code path (1):

            err_ret->text = "with Barry as BDFL, use '<>' "
                            "instead of '!='";

Later, it goes to code path (2) as well:

        if (tok->buf != NULL) {
            ...
            err_ret->text = (char *) PyObject_MALLOC(len + 1);

Hum, I modified my PR to removed *dead code*:

            err_ret->text = "with Barry as BDFL, use '<>' "
                            "instead of '!='";

This issue and bpo-34080 look unrelated to me. They can be fixed independently.

In practice, both issues are related and it seems easier to me to fix them both at the same time ;-)

AFAICT, that future only works in the REPL.

PR 8262 fixes the intended setting of error message for the "Barry as BDFL" easter egg.

>>> from __future__ import barry_as_FLUFL
>>> 1 != 2
  File "<stdin>", line 1
    1 != 2
       ^
SyntaxError: with Barry as BDFL, use '<>' instead of '!='

But there is other problem. As was exposed in the private communication with Victor and Barry, this easter egg works only in REPL (or if explicitly pass __future__.CO_FUTURE_BARRY_AS_BDFL to compile()). I'll try to fix this too.

The problem is that 'from __future__ import' is parsed after converting the sources to the AST. But this flag affects the behavior of the tokenizer, before an AST is created.

New changeset aba24ff by Serhiy Storchaka in branch 'master':
bpo-34084: Fix setting an error message for the "Barry as BDFL" easter egg. (GH-8262)
aba24ff

New changeset 220afff by Serhiy Storchaka (Miss Islington (bot)) in branch '3.7':
bpo-34084: Fix setting an error message for the "Barry as BDFL" easter egg. (GH-8262) (GH-8423)
220afff

New changeset ec729d5 by Serhiy Storchaka (Miss Islington (bot)) in branch '3.6':
bpo-34084: Fix setting an error message for the "Barry as BDFL" easter egg. (GH-8262) (GH-8424)
ec729d5