The bytecode compilers ignores ast.Str and ast.Num nodes:---------------------------->>> def func():...     123...     "test"... >>> import dis; dis.dis(func)  3           0 LOAD_CONST               0 (None)              3 RETURN_VALUE----------------------------But other ast nodes which are constant are not ignored:---------------------------->>> def func2():...     b'bytes'...     (1, 2)... >>> import dis; dis.dis(func2)  2           0 LOAD_CONST               1 (b'bytes')              3 POP_TOP  3           4 LOAD_CONST               4 ((1, 2))              7 POP_TOP              8 LOAD_CONST               0 (None)             11 RETURN_VALUE----------------------------I don't understand the point of loading a constant and then unload it (POP_TOP).Attached patch changes the compiler to not emit LOAD_CONST+POP_TOP anymore.My patch only affects constants. Example with the patch:---------------------------->>> def f():...  x... >>> import dis>>> dis.dis(f)  2           0 LOAD_GLOBAL              0 (x)              3 POP_TOP              4 LOAD_CONST               0 (None)              7 RETURN_VALUE----------------------------The compiler still emits "LOAD_GLOBAL x" for the instruction "x".Ignoring the Python instruction "x" would change the Python semantics, because the function would not raise a NameError anymore if x is not defined.Note: I noticed this inefficient bytecode while working on the issue#26146 (add ast.Constant).

The patch looks alright.Will the following code compile OK?  What will it compile to?   if 1:      42

LGTM.It looks to me that this optimization was added to avoid spending executing time for docstrings. Other cases almost never occur in real code and are not worth to be optimized. But the patch makes the code cleaner (it would even more cleaner if collapse all kinds of constants in Constant).

Serhiy: "It looks to me that this optimization was added to avoid spending executing time for docstrings."Hum, let me dig Mercurial history.----changeset:   39364:8ef3f8cf90e1branch:      legacy-trunkuser:        Neal Norwitz <nnorwitz@gmail.com>date:        Fri Aug 04 05:09:28 2006 +0000files:Lib/test/test_code.pyMisc/NEWSPython/compile.cdescription:Bug#1333982: string/number constants were inappropriately storedin the byte code and co_consts even if they were not used, ieimmediately popped off the stack.---https://bugs.python.org/issue1333982#msg26659:"For reference, an optimization that got lost: (...) 'a' is the docstring, but the 'b' previously did not show up anywhere in the code object.  Now there is the LOAD_CONST/POP_TOP pair."Ah, it was a regression introduced by the new AST compiler. But the change introduced a new optimization: numbers are now also ignored. In Python 2.4 (before AST), numbers were not ignored:--->>> def f():...  "a"...  1...  "b"... >>> import dis; dis.dis(f)  3           0 LOAD_CONST               1 (1)              3 POP_TOP                           4 LOAD_CONST               2 (None)              7 RETURN_VALUE        ---If we continue to dig deeper, before AST, I found:---changeset:   4991:8276916e1ea8branch:      legacy-trunkuser:        Guido van Rossum <guido@python.org>date:        Fri Jan 17 21:04:03 1997 +0000files:Python/compile.cdescription:Add co_stacksize field to codeobject structure, and stacksize argumentto PyCode_New() argument list.  Move MAXBLOCKS constant to conpile.h.Added accurate calculation of the actual stack size needed by thegenerated code.Also commented out all fprintf statements (except for a new one todiagnose stack underflow, and one in #ifdef'ed out code), and addedsome new TO DO suggestions (now that the stacksize is taken of the TODO list).---This patch added the following code to com_expr_stmt() inPython/compile.c:+       /* Forget it if we have just a doc string here */+       if (NCH(n) == 1 && get_rawdocstring(n) != NULL)+               return;I'm unable to find the exact part of the compiler which ignores strings in statement expressions.

"""Will the following code compile OK?  What will it compile to?   if 1:      42"""compile OK: what do you mean? This code doesn't make any sense to me :-)Currently text strings statements are ignored:--->>> def f():...  if 1:...   "abc"... >>> import dis>>> dis.dis(f)  3           0 LOAD_CONST               0 (None)              3 RETURN_VALUE---But byte strings emit a LOAD_CONST+POP_TOP:--->>> def g():...  if 1:...   b'bytes'... >>> dis.dis(g)  3           0 LOAD_CONST               1 (b'bytes')              3 POP_TOP              4 LOAD_CONST               0 (None)              7 RETURN_VALUE---With my patch, all constants statements will be ignored. Example with my patch:--->>> def h():...  if 1:...   b'bytes'... >>> import dis; dis.dis(h)  3           0 LOAD_CONST               0 (None)              3 RETURN_VALUE---

Serhiy: "It looks to me that this optimization was added to avoid spending executing time for docstrings. Other cases almost never occur in real code and are not worth to be optimized. But the patch makes the code cleaner (it would even more cleaner if collapse all kinds of constants in Constant)."Oh, I don't really care of performance. The bytecode just doesn't make any sense to me. I don't understand why we load a constant.Maybe the compiler should emit a warning to say that the code doesn't make sense at all and is ignored?Example with GCC:$ cat x.c int main(){    1;}$ gcc x.c -Wall -o xx.c: In function 'main':x.c:3:5: warning: statement with no effect [-Wunused-value]     1;     ^

> Maybe the compiler should emit a warning to say that the code doesn't make sense at all and is ignored?Oh ok, now I recall a similar issue that I posted 3 years ago: issue#17516, "Dead code should be removed".Example of suspicious code:def func():   func2(),func() calls func2() and then create a tuple of 1 item with the func2() result. See my changeset33bdd0a985b9 for examples in the Python source code. The parser or compiler should maybe help to developer to catch such strange code :-)In some cases, the code really contains code explicitly dead:def test_func():   return   do_real_stuff()do_real_stuff() is never called. Maybe it was a deliberate choice, maybe it was a mistake in a very old code base, bug introduced after multiple refactoring, and high turn over in a team? Again, we should emit a warning on such case?Hum, these warnings have a larger scope than this specific issue (don't emit LOAD_CONST for constants in expressions).See also the related thread on python-dev for the specific case of triple quoted strings ("""string"""):https://mail.python.org/pipermail/python-dev/2013-March/124925.htmlIt was admitted that it's a convenient way to insert a comment and it must not emit a warning (at least, not by default?)

+1

New changeseta0d053899ff8 by Victor Stinner in branch 'default':Simplify main() of test_asthttps://hg.python.org/cpython/rev/a0d053899ff8New changesetbcf27fa55632 by Victor Stinner in branch 'default':Replace noop constant statement with expressionhttps://hg.python.org/cpython/rev/bcf27fa55632

changeset:   100192:4bdb21380743tag:         tipuser:        Victor Stinner <victor.stinner@gmail.com>date:        Mon Feb 08 18:17:58 2016 +0100files:Lib/test/test_ast.pyLib/test/test_code.pyLib/test/test_grammar.pyMisc/NEWSPython/compile.cdescription:compiler now ignores constant statementsThe compile ignores constant statements and emit a SyntaxWarning warning.Don't emit the warning for string statement because triple quoted string is acommon syntax for multiline comments.Don't emit the warning on ellipis neither: 'def f(): ...' is a legit syntax forabstract functions.Changes:* test_ast: ignore SyntaxWarning when compiling test statements. Modify  test_load_const() to use assignment expressions rather than constant  expression.* test_code: add more kinds of constant statements, ignore SyntaxWarning when  testing that the compiler removes constant statements.* test_grammar: ignore SyntaxWarning on the statement "1"

I changed my patch to emit a SyntaxWarning. If too many users complain of the warning, maybe we can remove it. IMHO it's useful to detect bugs.

Shouldn't the message be "constant statement ignored"?  The current wording reads strange to me.

New changeset15531b10976c by Victor Stinner in branch 'default':compiler: don't emit SyntaxWarning on const stmthttps://hg.python.org/cpython/rev/15531b10976c

> Shouldn't the message be "constant statement ignored"?  The current wording reads strange to me.I removed the warning ;)

I think the warning was helpful; it just had confusing wording.Instead of: """>>> def f():...  False...<stdin>:2: SyntaxWarning: ignore constant statement"""perhaps: """>>> def f():...  False...<stdin>:2: SyntaxWarning: ignoring constant statement"""or even: """>>> def f():...  False...<stdin>:2: SyntaxWarning: ignoring unused constant 'False'"""

Sorry you are late :-) I started a thread on python-dev and it was decidedto let linters handle this warning.

FYI the thread was in February 2016:https://mail.python.org/pipermail/python-dev/2016-February/143163.html"[Python-Dev] Issue#26204: compiler now emits a SyntaxWarning on constant statement"