Probably there's a better example of tests that don't have to depend on the chainparam globals (in this case cGlobalSwitchingChainParams) anymore to make sure they test the same as when CBaseChainParams::MAIN is selected.
Note how the const for the params' reference wasn't possible before.

This looks dangerous, as it destroys the chainparams object created by the previous caller. Can you either document this, or switch to a solution that doesn't need that (for example, keeping a std::map<string, CChainParams)?

I don't like this change, it is useful to have the port numbers for both networks in the help output
(and we have the same for -port)

And I don't like special cases for testnet3 (or any other chain), which I usually find with rgrep "test" or rgrep CBaseChainParams::TESTNET.
Maybe we can loop through all the default ports for each supported chain as in
jtimon@28d961f ?
I mean, that would require something like jtimon@042fdb0 first.

As said, we're doing exactly the same for -port. What makes -rpcport different? #7128 unified these.

Looping through all networks would be overkill, there's no good reason to list regtest here, which is hidden (unless --help-debug). I think the current solution is fine.

@laanwj what about jtimon@dd35471 ?
This way you can bitcoind -testnet --help but bitcoind -regtest --help too.

re difference between -port and rpc-port, I'm removing the testnet3 special case from port too, see https://github.com/bitcoin/bitcoin/pull/6907/files#diff-c865a8939105e6350a50af02766291b7R374

rewarding unification, the change in port was also introduced only 5 days ago in #6961

This way you can bitcoind -testnet --help but bitcoind -regtest --help too.

I don't think that is better than just listing both in the standard help message. Parametrizing the help message on the network, I dunno, that's not something I'd expect as user.

That can be documented, but I'll just restore the old messages and make another local variable for testnet3.

well, 2 more.

Regarding looping and not wanting to show regtest's information, we could easily do that just by leaving regtest out of supportedChains (see jtimon@160d874#diff-64cbe1ad5465e13bc59ee8bb6f3de2e7R20 ). Although that would also leave it out from jtimon@160d874#diff-de33fb0129cd5121c7098035527b26fdR14 and jtimon@28d961f , which is probably not a big deal. The special case for testnet3 can be solved by replacing a "main and testnet" enumaration whether we loop around 2 or 3 elements.
Just a thought. Or we can just leave it as it is, it's just I thought that I had when I was paranoid about jtimon@28d961f maybe being controversial (as a "too altcoinish" commit) and I just remembered.
Of course, that can be done later if it's done.

Where is the !?

For some reason we we're using !testnet, which is the same as main and !!main, but the opposite of !!testnet.

I wanted to have solved it ages ago in jtimon@3d03f15 (#6423) but that commit depended on the cursed PR #6068 (just as cursed as its several predecessors).

In theory it's not defined for main, it's defined for testnet/regtest only;

For main, it is defined in https://github.com/bitcoin/bitcoin/blob/master/src/chainparams.cpp#L120

What that text refers to, is that users aren't currently allowed to -acceptnonstdtxn=true on mainnet, they can only set -acceptnonstdtxn=false in testnet/regtest, which makes the fact that the displayed default is false (when in testnet/regtest the default is actually true) even more stupid.
I wanted to have solved this in jtimon@a4b28c4 but it depends on the previously mentioned commit.

What about default: %u -> default: %u (testnet), default: %u (regtest)?

I really think we should solve this by allowing -acceptnonstdtxn=true on mainnet. See jtimon@a4b28c4