chiseltest for 5.1.0 in README

[kazutomo]

I'm somewhat curious about build.sbt in the README.
While the build.sbt example for 3.6.0 includes chiseltest with the comment "We also recommend using chiseltest for writing unit tests", the example for 5.1.0 does not include chiseltest.
Is this intentional, or was it perhaps an oversight? I assume chiseltest can be used for 5.1.0 (at least I use it).

[schoeberl]

There is an intention to move away from the current version of ChiselTest towards a new testing framework when Chisel switched to CIRCT (version 5). However, Kevin made ChiselTest available for Chisel 5.

[kazutomo]

Thank you for the insight! I was thinking that switching to CIRCT was a reason but was not 100% sure.

Now, an important question arises: What testing harness is recommended for Chisel 5.x or later?
svsim or external tool like cocotb?

I really like chiseltest. I guess people have tons of test benches using chiseltest. It would be very helpul if chiseltest (API) will be maintained in future release.

[seldridge]

svsim is the intended infrastructure on top of which things like a new ChiselTest could be built. Any Verilog-focused testing framework (like CocoTB) or manually written SV testbenches (which use new FIRRTL ABIs around probes) can always be used.

The main problem is that ChiselTest is tightly coupled to the Scala-based FIRRTL Compiler (SFC). The ChiselTest maintainers got ChiselTest to work with Chisel 5 by forking the SFC and republishing it. When you use ChiselTest with Chisel 5, it is recompiling your design using the SFC. This is much harder for Chisel 6 as Chisel 6 begins to use new FIRRTL constructs (probes, properties, etc.) which are both intended to make verification easier, but entirely unsupported in the SFC. These features could be supported at the cost of maintaining two compilers. The team working on Chisel and CIRCT can't justify this.

The best approach would be to port the core features of ChiselTest to work with svsim. This shouldn't be too hard.

[kazutomo]

Thank you for the internal details. I'm curious whether end users should use svsim directly or wait until ChiselTest to adopt svsim. I'm also curious whether formal test will be maintained in the future release.