CSV-285: Replace BufferedReader with PushbackReader

[belugabehr]

https://issues.apache.org/jira/browse/CSV-285

[belugabehr]

Hold off on this one for right now, the JMH tests are not as favorable as I was expecting. I need to review the tests.

[coveralls]

Coverage decreased (-0.3%) to 97.99% when pulling 532a3f1 on belugabehr:CSV-285 into 27843d8 on apache:master.

[garydgregory]

No rush. I might cut a release candidate next week or so but this is a major change we might only want to do after that.

[belugabehr]

Ya, all the unit tests pass, but I definitely see some flaws in my attempt here. Do not commit.

[belugabehr]

@garydgregory

Weird timing that I just happened to pick up CSV commons work the other day. I thought I was making big progress on performance, but my research just revealed that there was a big performance regression as part of adding support for String delimiters #76 which was just added to master recently. My contributions this week made some incremental improvements, but really the big performance improvement I addressed in #167 which bypasses the work in #76 by short-circuiting for the common use case of a single delimiter character. With #167 in place, the performance is much more similar to the current 1.8 version and not nearly as impressive as I had thought.

[garydgregory]

Hi @belugabehr,
Please note that I merged a bug fix today related to string delimiters.

[garydgregory]

Hi @belugabehr
Are you still working on this PR or should it be closed?

[belugabehr]

I'm going to close this.

I think a push-back buffer is cleaner as it's really well suited for this use case, but the current implementation is faster (and not all that hard to follow).