Erlang R12B3 的sub binary优化似乎有Bug？

阅读更多

09.3.18更新:
随着R13A release，我也重新查了一下这个问题，首先这份代码在编译的时候，会提示

引用

Warning: NOT OPTIMIZED: different control paths use different positions in the binary

litaocheng同学提醒，只要将代码修改为

dame_shit(Bin) ->
    dame_shit(Bin, <<"\r\n">>, not_found).
dame_shit(<>, Tag, not_found) ->
    dame_shit(T, Tag, found);
dame_shit(<>, Tag, not_found) ->
    dame_shit(T, Tag, not_found);
dame_shit(<<>>, _, _) ->
    not_found;
dame_shit(Bin, _, found) ->
    {found, Bin}.

编译就正常了:

引用

Warning: OPTIMIZED: creation of sub binary delayed

然后在R12B-5/R13A上面执行，逻辑正常：

引用

Find '\r\n' result1: {found,<<"something maybe wrong!">>}
Find '\r\n' result2: {found,<<"something maybe wrong!">>}

下面是原文
===========================
在使用过程中，发现了sub binary的优化似乎有Bug，先上代码：

-module(dame_shit).
-export([main/0, dame_shit/1, dame_shit2/1]).

dame_shit(Bin) ->
    dame_shit(Bin, <<"\r\n">>, not_found).
dame_shit(Bin, _, found) ->
    {found, Bin};
dame_shit(<>, Tag, not_found) ->
    dame_shit(T, Tag, found);
dame_shit(<>, Tag, not_found) ->
    dame_shit(T, Tag, not_found);
dame_shit(<<>>, _, _) ->
    not_found.

dame_shit2(Bin) ->
    dame_shit2(Bin, <<"\r\n">>, not_found).
dame_shit2(Bin, _, found) ->
    {found, Bin};
dame_shit2(<>=UnusedBinary, Tag, not_found) ->
    dame_shit2(T, Tag, found);
dame_shit2(<>, Tag, not_found) ->
    dame_shit2(T, Tag, not_found);
dame_shit2(<<>>, _, _) ->
    not_found.

main() ->
    io:format("Find '\\r\\n' result1: ~p~n", [dame_shit(<<"oops\r\nsomething maybe wrong!">>)]),
    io:format("Find '\\r\\n' result2: ~p~n", [dame_shit2(<<"oops\r\nsomething maybe wrong!">>)]),
    ok.

dame_shit & dame_shit2 的区别在于

<>=UnusedBinary

这句，这会导致binary优化失效，查看编译输出也可以发现这点：

引用

$ erlc +bin_opt_info dame_shit.erl
./dame_shit.erl:8: Warning: OPTIMIZED: creation of sub binary delayed
./dame_shit.erl:10: Warning: OPTIMIZED: creation of sub binary delayed
./dame_shit.erl:10: Warning: variable 'B' is unused
./dame_shit.erl:19: Warning: INFO: the '=' operator will prevent delayed sub bin
ary optimization
./dame_shit.erl:19: Warning: NOT OPTIMIZED: called function dame_shit2/3 does no
t begin with a suitable binary matching instruction
./dame_shit.erl:19: Warning: variable 'UnusedBinary' is unused
./dame_shit.erl:21: Warning: NOT OPTIMIZED: called function dame_shit2/3 does no
t begin with a suitable binary matching instruction
./dame_shit.erl:21: Warning: variable 'B' is unused

运行的时候确实是结果不一样：

引用

$ erl -noshell -s dame_shit main -s init stop
Find '\r\n' result1: not_found
Find '\r\n' result2: {found,<<"something maybe wrong!">>}

在
Windows XP + Erlang OTP R12B3
Linux X64_86 + Erlang OTP R12B3
下都得出相同的结果。


为什么会这样？在 dame_shit(<>, Tag, not_found) -> 下面加入日志输出：
io:format("B=~c~n", [B]),
执行后，输出如下：

引用

B=p
B=

B=m
B=h
B=g
B=a
B=e
B=r
B=g
B=!
Find '\r\n' result1: not_found

可见，每次执行这个匹配函数，指针都跳过了2个字符
oo ps\r \nso met hin g m ayb e w ron g!

跳过2个字符的原因，应该是由于前面的函数
dame_shit(<2/bytes, T/binary>>, Tag, not_found) ->
中的2bytes引起。