may 26 2022 discussion:
Need to add two tests:

• ipfs block get $(ipfs add --raw-leaves=false --chunker=size-$(( 1024 * 1024 )) 1MiBFile) | ipfs block put works. (non raw leaves blocks with 1MiB chunks include a ~10 bytes protobuf wrapping making them slightly too big, thoses blocks MUST pass, this is what this PR fixes)
• top down blackbox exchange test of 2MiB blocks.

Other possible solution if this drag on too much, revert 04e7e95 in 0.13.0 and move it to 0.14.0.

What's the priority of this? What is motivating this work?

What is motivating this work?

1. There is edge cases when you use 1MiB without --raw-leaves you exceed 1MiB because the protobuf wrapper.
2. linux2ipfs use 2MiB raw leaves blocks and I don't like being annoyed when I use it.

This still needs tests described in #8968 (review)

I know, I didn't took time to wrote them yet.

confirm two nodes are unable to exchange 2MiB+1 byte block

I didnt do this because we don't really have consensus on this: ipfs/specs#331 (comment)

Can we close this?

The number is still 1MiB in master:

Here is the repro:

hugo@rikus /tmp> truncate -s1GiB a
hugo@rikus /tmp> ipfs add --chunker=size-(math 1024\*1024) a
...
added Qmd7wspr7c4F6kAoaXqMiXsw6PBKGU3c7u8RAWMhuaF6vc a
hugo@rikus /tmp> ipfs dag export Qmd7wspr7c4F6kAoaXqMiXsw6PBKGU3c7u8RAWMhuaF6vc > a.car
...
hugo@rikus /tmp> ipfs dag import a.car 
Error: import failed at block "QmW5Wc6v9jMsvCDPJaFQxFmzprWthSL7getyav31oLwAeP": produced block is over 1MiB: big blocks can't be exchanged with other peers. consider using UnixFS for automatic chunking of bigger files, or pass --allow-big-block to override

See how a 1MiB chunk can't be imported.

Thanks @Jorropo and @gammazero for flagging. This is a bug we ned to fix before shipping IPIP-499 in 0.40.0-rc1.

This PR one is bit old, so created new PR #11185 with the limit raised to 2MiB on both kubo and boxo sides, plus boundary tests for block put, dag put, dag import, ipfs add (raw and wrapped leaves), and bitswap exchange between two nodes. Let's continue there.

Also added regression tests for the libp2p message size hard limit (~4MiB) to inform future spec evolution (+ catch any breaking change in go-libp2p, as it is happening very often recently)