We don't use all of the Stack functionality for it, so drop useless methods
and avoid some interface conversions. It increases single-node TPS by about
0.9%, so nothing really important there, but not a bad change either. Maybe it
can be reworked again with generics though.
v.estack is used throughout the code to work with estack, while ctx.sc.estack
is (theoretically) just a reference to it that is saved on script load and
restored to v.estack on context unload. The problem is that v.estack can grow
as we use it and can be reallocated away from its original slice (saved in the
ctx.sc.estack), so either ctx.sc.estack should be a pointer or we need to
ensure that it's correct when loading a new script. The second approach is a
bit safer for now and it fixes#2798.
Share parameters parsing code between 'contract invokefunction' and
'vm run' commands. It allows VM CLI to parse more complicated parameter
types including arrays and file-backed bytestrings.
Adding an array multiple times leads to the fast update via `IncRC`.
This hides the allocation that is there on the first addition. In this
commit add another benchmark which measures Add/Remove together, to
ensure that `switch` in `refCounter.Add` is entered. Benchmark results
are meaningful, because `Add`/`Remove` have almost identical implementation.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
And determine the need for Null dynamically. For some reason the only dynamic
context is Contract.Call. CALLT is not dynamic and neither is a call from
native contract, go figure...
We use it a lot in (*Blockchain).IsTxStillRelevant().
name old time/op new time/op delta
IsSignatureContract-8 19.1ns ± 5% 1.2ns ± 4% -93.81% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
IsSignatureContract-8 0.00B 0.00B ~ (all equal)
name old allocs/op new allocs/op delta
IsSignatureContract-8 0.00 0.00 ~ (all equal)
Perform add/set/remove operations with VM type firstly, and with
refcounter after that. It is needed to be able to extend add/set/remove
operations with extra checks.
We can omit DAO wrapping for safe methods and for those methods that are not
wrapped into try-catch block. However, we still need to persist
notificationsCount changes for these methods to the parent context.
Reference implementation:
e167f03e25/src/neo-vm/ExecutionContext.cs (L77).
Turns out that exception-related code (try\finally handling) is the only
place where next instruction pointer bounds are not pre-checked before
assignment. This statement is true for both go and C# implementations.
However, C# code has duplicate check located inside the
InstructionPointer setter for all instructions except the
exception-related ones. Neo-go code doesn't perform this duplicate check.
Thus, C# VM FAULTs execution in case of invalid TRY/FINALLY offsets, and
neo-go VM does not.
This commit reverts a part of
https://github.com/nspcc-dev/neo-go/pull/2396. This bug is discovered
thanks to test added in https://github.com/neo-project/neo-vm/pull/453.
Which allows to create verification scripts without keys.PublicKey which is
convenient in some cases where we already have serialized key and don't want
to waste time decompressing it.
Currently the only known reason this can happen is processing
ENDFINALLY opcode before the corresponding ENDTRY.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
As can be seen in https://dotnetfiddle.net/s7eg21 (int) conversions
result in an exception in C# code.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
Also do not limit depth. It was introduced in e34fa2e915 as a simple
solution to OOM problem. In this commit we do exactly the refactoring
described there. Maximum size is the same as stack item size and
can be changed if needed withouth significat refactoring.
`1 MiB` seems sufficient, though.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
When execution fails it's the only way to change the prompt showing the
last executed instruction. Example:
```
NEO-GO-VM > loadgo examples/engine/engine.go
READY: loaded 117 instructions
NEO-GO-VM 0 > run
Error: at instruction 3 (SYSCALL): failed to invoke syscall 805851437: syscall not found
NEO-GO-VM 8 > parse
Error: missing argument
NEO-GO-VM 8 > parse 123
Integer to Hex 7b
Integer to Base64 ew==
String to Hex 313233
String to Base64 MTIz
NEO-GO-VM 8 > unload
NEO-GO-VM > exit
```
These tests are slow on Windows, so refactor them a bit and avoid the following
error:
```
--- FAIL: TestRunWithDifferentArguments (4.01s)
cli_test.go:96:
Error Trace: cli_test.go:96
cli_test.go:321
Error: command took too long time
Test: TestRunWithDifferentArguments
```
For some reason `foo.go` is interpreted as an http URL, and even if we
replace it with `./foo.go` there is an errors with file missing on disk.
Because `CompileWithOptions` should be able to compile file under any
circumstances, we allocate temporary directory base on version used to
compile a binary.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
First argument contains filename, thus we use '.go' suffix to
distinguish between directories and files.
Contract name should be provided in options.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
It is a simple wrapper over `CompileWithOptions` which we don't really
need. Custom options can be specified explicitly instead of using some
random default. This default was introduced in 1578904da, however tests
written there use `CompileWithOptions` and all other tests
pass on that commit even without this default.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
When there is a single big instruction (like PUSHDATA4) in script,
all other instructions are padded to the right with spaces.
This makes it hard to view script in terminal, because long lines
are usually wrapped at the screen boundary and printed as multiple lines.
The culprit is our `cursor` field which is printed in the last column
and causes all previous fields to have the same length for every
instruction. One way to fix this is to omit cursor field if it is empty.
Signed-off-by: Evgeniy Stratonikov <evgeniy@nspcc.ru>
