Tracepoint format parser

Adds a new builtin "args", which acts as a pointer to the tracepoint
arguments struct when inside tracepoint probes.

fix string comparison

This patch fixes strcmp by reimplementing it as a IRBuilder method and
unrolling the loop based on the literal string size instead of
STRING_SIZE.

Fixes: https://github.com/iovisor/bpftrace/issues/14

truncate zero range on hist

fix BEGIN / END probes

add .gitignore

add verbose debug output with -dd

Docs

update usdt with arguments examples

BEGIN/END probes were broken after
https://github.com/iovisor/bpftrace/pull/100.

Fixes: https://github.com/iovisor/bpftrace/issues/51

Ignore common build directories. Useful to navigate the code base with
tools taking .gitignore into account.

LLVM optimizes IR before turning it into BPF instructions. Sometimes the
optimization might cut off important code if the IR was not built
correctly. -d is not enough to find these issues, therefore this patch
introduces more debug levels to also print the raw IR before it gets
optimized.

 usdt: resolve usdt arguments

Now we can infer the path from current_attach_point_, which means we
don't need to keep track of both references.

This patch introduces initial support for USDT arguments through the
argX builtins.  For example:
`./bpftrace -e 'usdt:bin:probe { printf("%s\n", str(arg0)); }'`

Ref: https://github.com/iovisor/bpftrace/issues/33

fix attach to tracepoint regression

Regression introduced by https://github.com/iovisor/bpftrace/pull/100.

We can attach to multiple tracepoints within the same process without
probem, which means we don't need to a suffix to it.

Fixes: https://github.com/iovisor/bpftrace/issues/103

add system() call

attach to multiple identical probes

There were two problems keeping bpftrace from attaching to multiple
identical probes in the same script: the first one was in our IR code
generation, where we were creating multiple sections with the same name
(since the secion name is determined by the probe's name). Since
sections must have unique names, LLVM was not able to translate our IR
to BPF. This was resolved by appending a `_#` to the section's name.

The second problem was attaching to that probe: we can't use the same
event name multiple times in "kprobe_events". bcc already handles
multiple probes coming from *different* processes, since it also
includes the PID on the event name. The solution was similar, we append
`_#` to the event name before sending it to bcc. This required some
strucutral changes, like keeping the `#` (index) in the AST Probe nodes
and calling `BPFTrace::add_probe` during codegen instead of during
semantic analysis (otherwise we wouldn't have the index for our Probes).

Fixes: https://github.com/iovisor/bpftrace/issues/97

More Tools

support negative integers