We'd switch to C++17, but we currently aim at supporting building revamb
with Ubuntu 16.04, whose default version of GCC doesn't support C++17.

This commit enforces on the whole project the usage of our own assertion
system. This means all calls to `abort`, `assert` and `llvm_unreachable`
have been replaced with calls to `revng_abort`, `revng_assert` and
`revng_unreachable`, respectively.

The error messages, usually expressed as `assert(Condition &&
"Message")` have now been replaced using the second (optional) argument
of `revng_assert`.

Additionally, all the `assert(false)` statements have been replaced with
calls to `revng_abort`. Apart from readibility, this ensures the
compilers is aware of the fact that call will never return.

This change will enable us to drop many statements whose sole purpose
was marking a variable employed in an assertion as used in release
mode. In fact, with the new assertion system this is no longer
necessary.

We used to have a `_` prefix for include guards macros, however,
according to the standard, macros starting with an underscore are
reserved. This commit drops them all.

This commit applies our `clang-format` rules globally. From now on, all
commits should respect our configuration file.

Pretty large change.

Include directives are groups by library, the groups should be sorted
from the most general (i.e., the STL) to the most specific (the local
includes).

This commit ensures that Boost includes are before LLVM includes.

This commit introduces `using` statements and `constexpr` functions to
increase readibility of various types.

Sometimes the true-branch of an `if` statement was using braces but the
false-branch was not. This commit fixes this.

This commit introduces an additional check in `check-conventions.sh` to
detect lines in source files exceeding 80 columns.

This commit introduces a check in `check-conventions.sh` to prevent
committing code denoted by a `// WIP` comment, that are supposed to be
used to remind the developer something has to be changed before
submitting a pull request.

For a long time we had an issue on ARM: the linker would use as the
first page 0x10000. However, on many distros, that matches the first
page that can be mapped (see `/proc/sys/vm/mmap_min_addr`). Therefore,
this shouldn't be a problem, except the fact that the translated program
also needs the preceeding page for the (outer) ELF header.

We used to suggest users to run:

    echo 4096 | sudo tee /proc/sys/vm/mmap_min_addr

Now in our tests, we force the text segment page to be at 0x20000,
therefore, this will no longer be necessary.

The symbol handling has been extended to register whether a symbol
represents a function or not. This information is then used to register,
during the global data harvesting phase, all the function symbols and
explicitly mark them through the "FunctionSymbol" `JTReason`.

We use this information during the CFEP harvesting phase to integrate
the information produced by the function boundaries detection with
potential unidentified CFEPs.

This option can be enabled with the `--use-debug-symbols`, which
supersedes `--use-sections`.

From version 0.25 pyelftools changed the API for ENUM_P_TYPE.
This commit wraps the import in a try-except block to handle older and
newer versions gracefully.

This problem has been fixed backporting the proper patches on LLVM.

This is a very large commit importing the reviewed (and heavily
simplified) stack analysis and the new ABI analysis, which provides
information on the calling convention of each function and so on.

For an overview of the new analyses please consult OVERVIEW.md.

This commit fixes a bug triggered by the fact that the constructor of
the `FilePortion` class was not initializing fields featuring a "native"
type (which are therefore not automatically zero-initialized).

In particular, this bug was triggered when dealing with a dynamic binary
lacking the sections for dynamic relocations.

`StreamWrapper` is a class that wraps a stream of any type. It has a
`flush` method, that, when called with a `std::stringstream` copies all
of its content in the wrapped stream.

The main reason for having `StreamWrapper` is being able to have a
`dumpInternal` method in a `.cpp` file while preserving a
stream-agnostic `dump` method in the header.

`Twine`s are very cool but very subtle objects. You basically have to
use them as `const Twine &` arguments only. In fact, a Twine keeps a
reference to objects whose lifetime corresponds to the current
statament. Therefore, a `Twine` can easily end up holding a reference to
invalid objects.

This commit fixes a bug in function isolation (which would show up only
if optimizations are enabled) due to a misusage of
`Twine`. Additionally, it reduces its usage to safe places in
`statistics.h`. Finally, we added an assertion in `check-conventions.sh`
to ensure that no variables of type `Twine` are ever declared.

revamb-dump was missing the `-T` option.

This commit simply reduces the length of lines in CMake by splitting the
statements over multiple lines. This is particularly useful when listing
the translation units composing a program/library. In fact, it makes
merge much easier.

`enum`s throw their entries into the parent scope. `enum class` can work
around this issue, however, `enum class` cannot have methods. Since we
often need to have `getName` and `fromName` functions for the enums, we
now create a namespace for each enum that contain a number of helper
functions.

Doxygen doc used to look for source files in the source root directory
only. We now use `git ls-files` to figure out which files need to be
part of the documentation.