support yaml tag preservation?

[clux]

python-yq supports "yaml roundtripping" and allows them to somehow preserve tags (which we support from #11)
by keeping track of them and re-injecting them back as yaml (when using -y as output).

python-yq setup

documented with https://kislyuk.github.io/yq/#preserving-tags-and-styles-using-the-y-yaml-roundtrip-option

they have this warning:

The -Y option is incompatible with jq filters that do not expect the extra information injected into the document to preserve the YAML formatting. For example, a jq filter that counts entries in the Instances array will come up with 4 entries instead of 2. A filter that expects all array entries to be mappings may break due to the presence of string metadata keys. Check your jq filter for compatibility/semantic validity when using the -Y option.
they seem to do it by keeping track of anchors and such in a dict via: https://github.com/kislyuk/yq/blob/ee87412fa89add226d263ec7885193e07243aa4c/yq/loader.py#L141-L155 and swapping back when serializing in https://github.com/kislyuk/yq/blob/ee87412fa89add226d263ec7885193e07243aa4c/yq/dumper.py#L33-L82 which seems super painful and hacky (internal __yq_alias__ keys for transport).

they also have the wildest regex based resolver system in there so not sure how viable this all is to a more closed down system like serde without writing your own serializer

Status

personally, don't particlarly care about this, but i also don't have use cases that need it atm.
if anyone has good ideas or want this. lemme know why or what. or post suggestions as to how to do it.

[evanfoster]

I just found this repo today after spending way too much time dealing with people having the different flavors of yq installed, and the ability to fully roundtrip is the main thing that would hold me up from migrating literally every script I have ever written to lq. I have a lot of YAML that's very heavily commented, and it would be fantastic to be able to preserve those comments. I don't really have much use for anchors, but preserving comments would be simply fantastic.

I'm very, very new with Rust, so I unfortunately have absolutely zero insights into how this might be achieved. I might take a stab at implementing it.

[clux]

hey thanks! this is definitely not easy, and am kind of in the process of getting rid of deprecated serde-yaml (which could not do this) in favor of serde-saphyr in #82. that change does not make it immediately possible. things i think we need;

1. serde-saphyr needs to be able to output what tags it found as part of deserialization
2. we need to keep track of them (and at what keys they were found)
3. we need to re-inject them before serialization
4. serde-saphyr needs to be able to be told about tags before serializing

possibly point 1 and 4 means we need to use the more lower level serde-saphyr-bw which has an actual Value (sequence variant contains anchors), but that may make the ser/de process quite complicated. i have not looked much at the bw layer.

investigation for this after #82 lands is definitely welcome though.