Software does not distinguish between different Conditions (Version: 1.5)

[cjcobb23]

I discovered some code that just doesn't smell right. I don't know the intended behavior, but I suspect the intent is different than what is actually happening.

The code that checks whether the condition that is required for a given RPC is met does not actually consider what condition is required. All of the conditions except NO_CONDITION are treated exactly the same. All of the following lines (in the same function) evaluate to the exact same value (false if required_condition is NO_CONDITION and true otherwise).
https://github.com/ripple/rippled/blob/3d86b49dae8173344b39deb75e53170a9b6c5284/src/ripple/rpc/impl/Handler.h#L80
https://github.com/ripple/rippled/blob/3d86b49dae8173344b39deb75e53170a9b6c5284/src/ripple/rpc/impl/Handler.h#L90
https://github.com/ripple/rippled/blob/3d86b49dae8173344b39deb75e53170a9b6c5284/src/ripple/rpc/impl/Handler.h#L91
https://github.com/ripple/rippled/blob/3d86b49dae8173344b39deb75e53170a9b6c5284/src/ripple/rpc/impl/Handler.h#L97
https://github.com/ripple/rippled/blob/3d86b49dae8173344b39deb75e53170a9b6c5284/src/ripple/rpc/impl/Handler.h#L117
The reason these lines all evaluate to the exact same value is because of the way these conditions are defined: all except NO_CONDITION have the least significant bit set to 1. So, when we bitwise and any two conditions together, the result is false if either are NO_CONDITION and true otherwise. I don't think this was the intended behavior of the code, since all of these lines could be replaced with condition_required != NO_CONDITION. Maybe someone who understands the intention of this piece of code can chime in here.

[cjcobb23]

For anyone scratching their head because git blame says I wrote this code, I actually just moved it from a different file. Prior to commit 7d867b806d70fc41fb45e3e61b719397033b272c, the code was here:
https://github.com/ripple/rippled/blob/761bb5744e9a43ebc6c94441edf8b2d8b1d662a4/src/ripple/rpc/impl/RPCHandler.cpp#L152

[nbougalis]

D'oh... this code is ancient (probably around 6 years old) and it does have the bug you describe, which is incredibly subtle.

I suspect the intention was to communicate that, e.g. NEEDS_CURRENT_LEDGER requires NEEDS_NETWORK_CONNECTION but all it did was make the code brittle. We should reconsider how this is handled, so that the condition checking code is both type-safe and robust by design.

In the meantime, your proposed solution is correct: replace the condition_required & WHATEVER statements with ((condition_required & WHATEVER) == WHATEVER). This retains the brittleness, but at least now we know about it.

Alternatively, we can change the enum values to be just powers of two, and change:

if ((condition_required & NEEDS_NETWORK_CONNECTION) &&
    (context.netOps.getOperatingMode() < OperatingMode::SYNCING))

to this:

// If any conditions are imposed, a network connection is required
if ((condition_required != NO_CONDITION) &&
    (context.netOps.getOperatingMode() < OperatingMode::SYNCING))

I think that this is minimal and probably less brittle.

[Coderayush13]

Hey I want to work on these isssue

[intelliot]

@Coderayush13 or anyone else: please feel free to work on this issue. Comment here with any questions you have.