Add verbatim builtin.

[chriseth]

Fixes #10869

This seems to work end-to-end.
Needs:

• tests
• figure out how the argument can be conveniently encoded in hex instead of a regular ascii string.

[chriseth]

We might need to make the side-effects a bit worse, by the way: This could use msize and so on.

[mijovic]

I think Changelog item should be added.
If verbatim can be used in inline assembly block, we should add such test.
Also, can we add some simple semantic test with inline assembly, of course with opcodes known to evm?

[chriseth]

Added the check to the optimizer test.

[chriseth]

@axic added empty bytecode check: https://github.com/ethereum/solidity/pull/11123/files#diff-8d0720c38c568d9c79f516230222679d874ba33c7361174ad052f275fc11a874R382

[axic]

Had to re-read the actual source code (not this PR). It seems for the other builtins, we have reserved them in Yul entirely, not just for the object version.

If we do it here, are we afraid some people use verbatim* in inline assembly? We should aim to reserve it in 0.9.0 at least, if so. (Or perhaps by that time verbatim will be supported in inline assembly?)

[chriseth]

Sounds good!

[axic]

I'm confused, what is the verdict? We lave it as-is and change it in 0.9.0?

[chriseth]

Yes!

[chriseth]

Re-checked the EVMASM optimizer:

• Inliner is unaffected since it ignores all blocks that breaksCSEAnalysisBlock.
• Jumpdest remover is unaffected since it removes unreferenced jumpdests, but verbatim is not supposed to jump to an actual jumpdest
• peephole optimizer is unaffected. It might remove the verbatim if it is unreachable, documented that.
• block deduplicator can combine identical verbatim bytecode blocks. Documented that.
• CSE splits at verbatim blocks and thus is unaffected.
• Constant Optimizer is unaffected.

[hrkrshnn]

A test to print the side effects test/libyul/functionSideEffects/verbatim.yul throws.

{
    function f() { verbatim_0i_0o("test") }
}

Here is a stack trace:

#0  0x00007ffff5ffb492 in __cxxabiv1::__cxa_throw (obj=0x1530910, tinfo=0x14e9788 <typeinfo for std::out_of_range>, dest=0x7ffff6011070 <std::out_of_range::~out_of_range()>)
    at ../../../../libstdc++-v3/libsupc++/eh_throw.cc:78
#1  0x00007ffff5ff23f8 in std::__throw_out_of_range (__s=0x102c79e "map::at") at ../../../../../libstdc++-v3/src/c++11/functexcept.cc:82
#2  0x0000000000d13d6f in std::map<solidity::yul::YulString, std::set<solidity::yul::YulString, std::less<solidity::yul::YulString>, std::allocator<solidity::yul::YulString> >, std::less<solidity::yul::YulString>, std::allocator<std::pair<solidity::yul::YulString const, std::set<solidity::yul::YulString, std::less<solidity::yul::YulString>, std::allocator<solidity::yul::YulString> > > > >::at (this=0x7fffffffd820, __k=...) at /usr/include/c++/10/bits/stl_map.h:550
#3  0x0000000000d73c19 in operator()<solidity::yul::SideEffectsPropagator::sideEffects(const solidity::yul::Dialect&, const solidity::yul::CallGraph&)::<lambda(solidity::yul::YulString, auto:24&&)>&>(solidity::yul::YulString, struct {...} &) (__closure=0x7fffffffd700, _function=..., _recurse=...)
    at /home/hari/s/r2/libyul/optimiser/Semantics.cpp:139
#4  0x0000000000d73ffe in solidity::yul::SideEffectsPropagator::sideEffects (_dialect=..., _directCallGraph=...) at /home/hari/s/r2/libyul/optimiser/Semantics.cpp:144
#5  0x00000000005384e6 in solidity::yul::test::FunctionSideEffects::run (this=0x1532b30, _stream=..., _linePrefix="  ", _formatted=true)
    at /home/hari/s/r2/test/libyul/FunctionSideEffects.cpp:94
#6  0x0000000000437bf4 in TestTool::process (this=0x7fffffffde00) at /home/hari/s/r2/test/tools/isoltest.cpp:172
#7  0x00000000004392ef in TestTool::processPath (

The problem is https://github.com/ethereum/solidity/blob/verbatimdata/libyul/optimiser/Semantics.cpp#L139. Not sure why it should get to that branch. Also don't have a direct example that throws. (Turns out that m_objectAccess is false and m_dialect->builtin("verbatim_...") returns nullptr. Maybe only an issue during testing?)

[chriseth]

@hrkrshnn what is the command you run?

[chriseth]

ah, `functionSideEffects' in isoltest?

[chriseth]

Depends on #11292

[hrkrshnn]

Added a couple of tests #11316

[axic]

Oh, this got increased to three members? I think it should really be documented what they are, I assume args,rets,data?

[chriseth]

Two lines up?