passing the arg through stdin instead of as command line arg for dfx canister call

[luc-mercier]

Sometimes passing the argument as commend line flag is not so convenient.

Two cases are:

• Canisters that take in raw binary data that is not Candid, and the caller wants to call some function to encode it. Using $(...) is not always easy. In the example of the registry canister, while I know I can encode in command line an argument, I never succeeded in wrapping it inside a $(...), the shell is doing something that I don't want.

• Reading from a file, when the argument is large. This applies to both raw binary and candid text representation.

I suggest that there is some flag to read the arg from stdin.

[chenyan-dfinity]

Just curious, which canister is not using Candid and why?

[luc-mercier]

The registry is not using Candid, for instance.

Why: I wasn't involved in the decision, I'm not sure.

But I'm pretty sure that as long as it's allowed to use arbitrary bytes for their interfaces, some developers will do so.

[nomeata]

A --raw flag is also useful to test Candid canisters with invalid or tricky Candid values, so this is reasonable.

hfx call supports --raw-arg as an alternative to --candid, but admittedly it is a more low-level tool.

[luc-mercier]

@nomeata I actually have the same FR for hfx: with hfx call --raw-arg, I'm also hitting the same problem: using $(...) to evaluate the --raw-arg causes difficulties with the shell. I've never managed to pass a binary proto to hfx call for instance, despite having confirmation that my command inside the $(...) does what it should.

[nomeata]

Right, reading from a file or stdin is a useful feature, but it’s kinda orthogonal to Candid vs. Raw, isn’t it?

[luc-mercier]

Right, reading from a file or stdin is a useful feature, but it’s kinda orthogonal to Candid vs. Raw, isn’t it?

Oh, absolutely. It was just to give the motivation for this request.

[nomeata]

Is this a “nice to have” or a “this would make my life better on a regular basis” for you?

[luc-mercier]

I'm actually not sure.

I like to do manual testing of code I write in addition to automated testing. With e.g. the registry canister, I have NEVER succeeded in calling it by hand (purely because of this issue, I believe). However:

• automated testing works fine, I have added a e2e test for the registry, and that was much easier than trying to do manual testing
• for newer NNS canisters we've started using Candid, for which I should be able to pass the arg as text (not binary) and not have this issue.

So right now this PR is not a high priority for me. But I suspect the need will come again occasionally, maybe not during "pure coding" phases, but more in "let's make sure everything really works" or "debug stuff in prod" phases.

[nomeata]

Well, let us know if the pain becomes bigger!

[ericswanson-dfinity]

This is now available through the --argument-file parameter, implemented in #2328

Example:

$ echo '("XYZ")' | dfx canister call dd_backend greet --argument-file -
("Hello, XYZ!")