Added support for building client with readline (bug1266386)

[tplavcic]

This change is to enable building mysql client with readline library since upstream mysql switched to editline and there were multiple requests to enable building with readline again (bug1266386).

Build with system readline is here set as default but it can be enforced with -DWITH_READLINE=system
Basically options for building are:
-DWITH_EDITLINE=system|bundled or -DWITH_READLINE=system
And system readline is the default.

Here is the test build:
rpm: http://jenkins.percona.com/job/percona-server-5.6-redhat-binary/143/
deb: http://jenkins.percona.com/job/percona-server-5.6-debian-binary/137/
deb32: http://jenkins.percona.com/job/percona-server-5.6-debian-binary-32/122/
bin: http://jenkins.percona.com/job/percona-server-5.6-binaries-release/107/

Here is the param build:
http://jenkins.percona.com/job/percona-server-5.6-param/837/

Some testing:
readline link in mysql client lib
[vagrant@t-centos6-64 ~]$ ldd /usr/bin/mysql
...
libreadline.so.6 => /lib64/libreadline.so.6 (0x00007fe657bba000)

Running:
mysql> show variables;
Ctrl+w
mysql> show
Comment: in editline Ctrl+w deletes the whole line, while in readline it deletes the characters from pointer back to first space character.

Reverse search:
In readline:
Ctrl+r
(reverse-i-search)`sh': show engines;

In editline it looks like this:
mysql> show engines;
bck:show

[laurynas-biveinis]

Why?

[tplavcic]

That's not a change, it's part of the centos7 patch so github gui is marking it red/green since it contains +/- start of line.

[laurynas-biveinis]

Did you test -DWITH_EDITLINE=system and bundled with this patch?

What happens if both WITH_EDITLINE and WITH_READLINE are specified?

Do completion and history across the sessions functions work in the
new client binaries? The reason I'm asking is because the fix seems to
be different from the WebScaleSQL one:
facebookarchive/webscalesql-5.6@4136922#diff-af3b638bc2a3e6c650974192a53c7291.
Removing libedit support aside, the differences are:

• they link with curses. They are not alone, a patch contributed on
  http://bugs.mysql.com/bug.php?id=63130 says "(...) does not handle the
  case where the Readline library uses termcap functions, but does not
  link with the termcap or curses library itself." Do you have any
  idea why it's not needed for us then?
• they require readline/history.h while we appear not to and trust the
  extern "C" declarations in mysql.cc to substitute? Replacing those
  declarations with an include seems more robust to me.
• ...and maybe required there and in CMake tests as it is history.h
  which defines HIST_ENTRY, and so, under Ubuntu 14.10 we can see now

-- Performing Test READLINE_HAVE_HIST_ENTRY
-- Performing Test READLINE_HAVE_HIST_ENTRY - Failed

because only readline.h but not history.h is included.
This is also mentioned in the description of #63130.

Once this goes in, this might fix more bugs than 1266386 alone, please
follow the links from that bug report:

• http://bugs.mysql.com/bug.php?id=63130
• https://bugs.launchpad.net/percona-server/+bug/1296192 / http://bugs.mysql.com/bug.php?id=72108
• http://bugs.mysql.com/bug.php?id=69991
• http://bugs.mysql.com/bug.php?id=68925
• https://bugs.launchpad.net/percona-server/+bug/1332822

It should be verified that readline binaries fix the issues in the bug
reports, and the bugs updated accordingly. The upstream bugs that do
not have LP bugs can be mentioned in 1296192, they all seem to be
caused by history file incompatibility.

[tplavcic]

I have tested building with -DWITH_EDITLINE=system and bundled previously and now with this latest commit and didn't have issues - I think I only tested running the client with previous patch with editline.

In the new commit I have added that if you specify both build options it exits with error.
Looks like this:
CMake Error at cmake/readline.cmake:261 (MESSAGE):
Cannot configure WITH_READLINE and WITH_EDITLINE! Use only one setting.

We link with curses also - variable ${MY_READLINE_LIBRARY} includes ${READLINE_LIBRARY} and ${CURSES_LIBRARY}.

In the second commit I changed the stuff you wrote about including the history.h stuff and extern functions. I tried something like this before but had some issues in some conditions so left it like it was - but that time I didn't see the patch you mentioned - so it seems it works.

From the bugs listed above the issues were:
history file format
now looks like this:
vagrant@t-debian7-64:~$ cat .mysql_history
show tables;
use test;
show tables;
show engines;
quit;

putty command line mixup when resizeing windows
I don't have proof for this and I only tried it with putty on linux (don't have windows).
Basically I tried to repeat the steps from bug: https://bugs.launchpad.net/percona-server/+bug/1332822
but it works ok for me.

general readline hotkey usage
Runs as described in my first comment.

I'll update the bugs once we agree that this is ok and it gets merged.

BUILDS:
build:
rpm: http://jenkins.percona.com/job/percona-server-5.6-redhat-binary/144/
deb: http://jenkins.percona.com/job/percona-server-5.6-debian-binary/138/
deb-32: http://jenkins.percona.com/job/percona-server-5.6-debian-binary-32/123/
bin: http://jenkins.percona.com/job/percona-server-5.6-debian-binary-32/123/

param:
http://jenkins.percona.com/job/percona-server-5.6-param/843/
I have incidently stopped the trusty-debug build while trying to do a new run with the commit I have later removed. If you think it's needed I can rerun it - although the one below is runing with that removed commit (which is practically the same thing).
http://jenkins.percona.com/job/percona-server-5.6-param/844/

Linking looks like this on debian7:
vagrant@t-debian7-64:~$ ldd /usr/bin/mysql
linux-vdso.so.1 => (0x00007fff921ff000)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fdee17f1000)
libreadline.so.6 => /lib/x86_64-linux-gnu/libreadline.so.6 (0x00007fdee15aa000)
libncurses.so.5 => /lib/x86_64-linux-gnu/libncurses.so.5 (0x00007fdee1387000)
libtinfo.so.5 => /lib/x86_64-linux-gnu/libtinfo.so.5 (0x00007fdee115e000)
libz.so.1 => /lib/x86_64-linux-gnu/libz.so.1 (0x00007fdee0f47000)
librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007fdee0d3e000)
libssl.so.1.0.0 => /usr/lib/x86_64-linux-gnu/libssl.so.1.0.0 (0x00007fdee0ade000)
libcrypto.so.1.0.0 => /usr/lib/x86_64-linux-gnu/libcrypto.so.1.0.0 (0x00007fdee06fa000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fdee04f5000)
libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fdee01ee000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fdedff6c000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fdedfd55000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fdedf9cb000)
/lib64/ld-linux-x86-64.so.2 (0x00007fdee1a14000)

CMAKE:
08:38:09 -- Found Curses: /usr/lib/x86_64-linux-gnu/libcurses.so
08:38:09 -- Looking for tputs in /usr/lib/x86_64-linux-gnu/libcurses.so
08:38:09 -- Looking for tputs in /usr/lib/x86_64-linux-gnu/libcurses.so - found
08:38:09 -- Looking for 3 include files stdio.h, ..., readline/history.h
08:38:09 -- Looking for 3 include files stdio.h, ..., readline/history.h - found
08:38:09 -- READLINE_INCLUDE_DIR /usr/include/readline
08:38:09 -- READLINE_LIBRARY /usr/lib/x86_64-linux-gnu/libreadline.so
08:38:09 -- Performing Test READLINE_HAVE_HIST_ENTRY
08:38:09 -- Performing Test READLINE_HAVE_HIST_ENTRY - Success
08:38:09 -- Performing Test READLINE_USE_LIBEDIT_INTERFACE
08:38:09 -- Performing Test READLINE_USE_LIBEDIT_INTERFACE - Success
08:38:09 -- Performing Test READLINE_USE_NEW_READLINE_INTERFACE
08:38:09 -- Performing Test READLINE_USE_NEW_READLINE_INTERFACE - Success

[laurynas-biveinis]

Could you please flatten the two commits (i.e. git rebase -i) and then it's OK to merge

[laurynas-biveinis]

Thanks Tomislav, please update the bugs as well

[mal]

Hi @tplavcic would you consider submitting this to upstream https://github.com/mysql/mysql-server?

I've successfully used your patch to restore sane readline functionality to my personal mysql client (my vim addled fingers thank you!), and there appears to be an open issue for doing just this: http://bugs.mysql.com/bug.php?id=69991

[tplavcic]

Hi @mal I would consider it and will put it into my todo list but I make no further promises! IIRC this patch doesn't break the current editline builds so they might even consider it.
Cheers!

[mal]

That's all I could ask for, thanks!