Minimize API surface in anticipation of gorelease

[johanbrandhorst]

The grpc-gateway was created long before things like go modules semver and with little concern for internal package management. This unfortunately means that technically the API surface is enormous, however most of these packages are not really considered part of the public API.

This change moves everything that isn't part of the public API under various internal folder structures. This is obviously technically a breaking change, but I'm hoping we can get away with saying that our users really shouldn't have been importing these things anyway if any of them have been. I'm happy to discuss any concerns.

The end goal of this migration is to be able to use https://godoc.org/golang.org/x/exp/cmd/gorelease to ensure API stability across minor patches. The current structure prohibits this (because we make breaking changes to the examples all the time).

CC @achew22 @tmc @ivucica

[achew22]

I'm extremely supportive of this change. Thank you for taking the initiative

[ivucica]

I am basically ok with this. Some tests are failing. I'll try to review this some time this week. Poke me using some out-of-band channel if I forget about it.

[codecov-io]

Codecov Report

Merging #1146 into master will decrease coverage by 0.04%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #1146      +/-   ##
==========================================
- Coverage   53.78%   53.74%   -0.05%     
==========================================
  Files          42       42              
  Lines        4211     4211              
==========================================
- Hits         2265     2263       -2     
- Misses       1698     1701       +3     
+ Partials      248      247       -1

Impacted Files	Coverage Δ
examples/internal/server/flow_combination.go	0% <ø> (ø)
examples/internal/server/unannotatedecho.go	0% <ø> (ø)
examples/internal/server/responsebody.go	0% <ø> (ø)
examples/internal/server/a_bit_of_everything.go	0% <ø> (ø)
examples/internal/server/non_standard_names.go	0% <ø> (ø)
internal/descriptor/grpc_api_service.go	0% <ø> (ø)
...-gen-grpc-gateway/internal/gengateway/generator.go	38.61% <ø> (ø)
protoc-gen-swagger/internal/genswagger/types.go	0% <ø> (ø)
internal/httprule/parse.go	86.32% <ø> (ø)
examples/internal/server/main.go	0% <ø> (ø)
... and 15 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update df4ff8d...7e41527. Read the comment docs.

[johanbrandhorst]

There seems to be a consensus that we should try this and see if it angers any users. Here goes!

[karelbilek]

why did not grpc-gateway bump version to v2 when doing this? It did break backward compatibility and one of our modules broke, because it imported httprule.

this is why v2 exists. grpc-gateway should have bumped that :(

I agree with the reason of doing this (reducing the public API), but it should change major version and thus import path when doing it

[karelbilek]

see https://blog.golang.org/v2-go-modules and https://github.com/golang/go/wiki/Modules - modules should be backwards compatible or change import path

I realize this is why this PR exists in the first place (to be able to track breaking changes), but it is still a massive breaking change in itself :(

[johanbrandhorst]

Hi @karel-3d, in principle I agree with you, but we wanted to see how much we could get away with doing without bumping to v2 yet. If you have a dependency that depends on httprule, I can move it back into the public API and make a new release (see #1154 and #1161). Could you raise an issue?

[karelbilek]

@johanbrandhorst hey sorry for the little confrontational tone at first, it was a bit angry at first :) but makes sense, and it's not that popular import so I get it

anyway I made an issue

#1168