summaryrefslogtreecommitdiffhomepage
path: root/src/numeric.c
diff options
context:
space:
mode:
authorRyan Lopopolo <[email protected]>2019-06-22 12:07:21 +0100
committerRyan Lopopolo <[email protected]>2019-06-22 12:31:01 +0100
commit0f516bbe1dd61759faf2609970bd5cd855931221 (patch)
tree68c725199ff2c04d40789ce9b2609f67ab60fe36 /src/numeric.c
parentf94551ad40c12370770c8e743ba8b39abc430547 (diff)
downloadmruby-0f516bbe1dd61759faf2609970bd5cd855931221.tar.gz
mruby-0f516bbe1dd61759faf2609970bd5cd855931221.zip
Add paragraph mode to String#each_line in mrblib
mruby/mruby#4511 demonstrated an infinite loop in `String#each_line` when given an empty string separator. In MRI, an empty separator places String#each_line in paragraph mode, where the String is separated on successive runs of newlines. In paragraph mode, the String `"abc\n\n\ndef\nxyz"` is split into `["abc\n\n\n", "def\nxyz"]`. This commit makes the String#each_line implementation as close to ruby/spec compliant as possible given the limitations of mruby core. With this patch, the following specs fail for `String#each_line`: - uses `$/` as the separator when none is given (can be fixed by aliasing and redefining the method to use $/ as the default value of separator in mruby-io) - when no block is given returned Enumerator size should return nil (`Enumerator#size` is not supported on mruby) - tries to convert the separator to a string using to_str (`String#to_str` is not implemented on mruby) This patch has similar memory consumption compared to the prior implementation and is takes 4x the time the prior implementation takes to execute: ```console /usr/bin/time -l ./bin/mruby -e '("aaa\n\nbbbbb\n\n\n\n\ncccc" * 100000).each_line("\n") { }'; ```
Diffstat (limited to 'src/numeric.c')
0 files changed, 0 insertions, 0 deletions