Change sub-types to TypeID instead of full Type redefinition

[grantseltzer]

What does this PR do?

Flattens the type structure of the IR to have TypeID for referencing of other types within things like struct fields, or pointed-to-types.

Motivation

While writing decoding code I realized that i'd have to redefine the Type that already exists in the Type field of Program.

Describe how you validated your changes

Forthcoming discussion with team.

[ajwerner]

Can you provide an example of where this is more useful or convenient? We had all ID based references before, and the general feeling was that this connected graph would be easier to use — it means you don’t need to go back and consult a data structure for every lookup

[cit-pr-commenter]

Regression Detector

Regression Detector Results

Metrics dashboard
Target profiles
Run ID: dbea650b-f1ce-4827-91a4-000df4de3b42

Baseline: dfcd7cc
Comparison: 3fe54ff
Diff

Optimization Goals: ✅ No significant changes detected

Fine details of change detection per experiment

perf	experiment	goal	Δ mean %	Δ mean % CI	trials	links
➖	docker_containers_cpu	% cpu utilization	+3.81	[+0.54, +7.08]	1	Logs
➖	tcp_syslog_to_blackhole	ingress throughput	+1.39	[+1.33, +1.45]	1	Logs
➖	file_tree	memory utilization	+1.38	[+1.22, +1.54]	1	Logs
➖	uds_dogstatsd_to_api_cpu	% cpu utilization	+0.85	[-0.03, +1.73]	1	Logs
➖	docker_containers_memory	memory utilization	+0.22	[+0.14, +0.30]	1	Logs
➖	quality_gate_idle_all_features	memory utilization	+0.10	[+0.01, +0.19]	1	Logs bounds checks dashboard
➖	ddot_logs	memory utilization	+0.10	[-0.05, +0.24]	1	Logs
➖	file_to_blackhole_1000ms_latency	egress throughput	+0.03	[-0.61, +0.66]	1	Logs
➖	file_to_blackhole_100ms_latency	egress throughput	+0.02	[-0.64, +0.68]	1	Logs
➖	file_to_blackhole_0ms_latency_http2	egress throughput	+0.02	[-0.60, +0.63]	1	Logs
➖	file_to_blackhole_0ms_latency_http1	egress throughput	+0.01	[-0.61, +0.62]	1	Logs
➖	uds_dogstatsd_to_api	ingress throughput	+0.00	[-0.22, +0.23]	1	Logs
➖	ddot_metrics	memory utilization	-0.00	[-0.12, +0.11]	1	Logs
➖	tcp_dd_logs_filter_exclude	ingress throughput	-0.01	[-0.03, +0.01]	1	Logs
➖	file_to_blackhole_300ms_latency	egress throughput	-0.02	[-0.66, +0.62]	1	Logs
➖	file_to_blackhole_500ms_latency	egress throughput	-0.04	[-0.61, +0.54]	1	Logs
➖	file_to_blackhole_1000ms_latency_linear_load	egress throughput	-0.04	[-0.28, +0.20]	1	Logs
➖	file_to_blackhole_0ms_latency	egress throughput	-0.05	[-0.69, +0.59]	1	Logs
➖	uds_dogstatsd_20mb_12k_contexts_20_senders	memory utilization	-0.10	[-0.14, -0.06]	1	Logs
➖	otlp_ingest_metrics	memory utilization	-0.17	[-0.33, -0.01]	1	Logs
➖	quality_gate_idle	memory utilization	-0.26	[-0.33, -0.19]	1	Logs bounds checks dashboard
➖	otlp_ingest_logs	memory utilization	-0.59	[-0.72, -0.46]	1	Logs
➖	quality_gate_logs	% cpu utilization	-1.53	[-4.26, +1.20]	1	Logs bounds checks dashboard

Bounds Checks: ❌ Failed

perf	experiment	bounds_check_name	replicates_passed	links
❌	docker_containers_memory	memory_usage	0/10
✅	docker_containers_cpu	simple_check_run	10/10
✅	docker_containers_memory	simple_check_run	10/10
✅	file_to_blackhole_0ms_latency	lost_bytes	10/10
✅	file_to_blackhole_0ms_latency	memory_usage	10/10
✅	file_to_blackhole_0ms_latency_http1	lost_bytes	10/10
✅	file_to_blackhole_0ms_latency_http1	memory_usage	10/10
✅	file_to_blackhole_0ms_latency_http2	lost_bytes	10/10
✅	file_to_blackhole_0ms_latency_http2	memory_usage	10/10
✅	file_to_blackhole_1000ms_latency	memory_usage	10/10
✅	file_to_blackhole_1000ms_latency_linear_load	memory_usage	10/10
✅	file_to_blackhole_100ms_latency	lost_bytes	10/10
✅	file_to_blackhole_100ms_latency	memory_usage	10/10
✅	file_to_blackhole_300ms_latency	lost_bytes	10/10
✅	file_to_blackhole_300ms_latency	memory_usage	10/10
✅	file_to_blackhole_500ms_latency	lost_bytes	10/10
✅	file_to_blackhole_500ms_latency	memory_usage	10/10
✅	quality_gate_idle	intake_connections	10/10	bounds checks dashboard
✅	quality_gate_idle	memory_usage	10/10	bounds checks dashboard
✅	quality_gate_idle_all_features	intake_connections	10/10	bounds checks dashboard
✅	quality_gate_idle_all_features	memory_usage	10/10	bounds checks dashboard
✅	quality_gate_logs	intake_connections	10/10	bounds checks dashboard
✅	quality_gate_logs	lost_bytes	10/10	bounds checks dashboard
✅	quality_gate_logs	memory_usage	10/10	bounds checks dashboard

Explanation

Confidence level: 90.00%
Effect size tolerance: |Δ mean %| ≥ 5.00%

Performance changes are noted in the perf column of each table:

• ✅ = significantly better comparison variant performance
• ❌ = significantly worse comparison variant performance
• ➖ = no significant change in performance

A regression test is an A/B test of target performance in a repeatable rig, where "performance" is measured as "comparison variant minus baseline variant" for an optimization goal (e.g., ingress throughput). Due to intrinsic variability in measuring that goal, we can only estimate its mean value for each experiment; we report uncertainty in that value as a 90.00% confidence interval denoted "Δ mean % CI".

For each experiment, we decide whether a change in performance is a "regression" -- a change worth investigating further -- if all of the following criteria are true:

1. Its estimated |Δ mean %| ≥ 5.00%, indicating the change is big enough to merit a closer look.

2. Its 90.00% confidence interval "Δ mean % CI" does not contain zero, indicating that if our statistical model is accurate, there is at least a 90.00% chance there is a difference in performance between baseline and comparison variants.

3. Its configuration does not mark it "erratic".

CI Pass/Fail Decision

✅ Passed. All Quality Gates passed.

• quality_gate_idle_all_features, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_idle_all_features, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_idle, bounds check intake_connections: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check memory_usage: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check lost_bytes: 10/10 replicas passed. Gate passed.
• quality_gate_logs, bounds check intake_connections: 10/10 replicas passed. Gate passed.

[agent-platform-auto-pr]

Static quality checks

✅ Please find below the results from static quality gates
Comparison made with ancestor dfcd7cc

Successful checks

Info

	Quality gate	Delta	On disk size (MiB)	Delta	On wire size (MiB)
✅	agent_deb_amd64	$${\color{green}-0}$$	$${\color{green}747.05}$$ < $${752.99}$$	$${\color{green}-0.01}$$	$${\color{green}185.0}$$ < $${187.44}$$
✅	agent_deb_amd64_fips	$${\color{green}0}$$	$${\color{green}745.43}$$ < $${751.36}$$	$${\color{green}-0.01}$$	$${\color{green}184.39}$$ < $${187.06}$$
✅	agent_heroku_amd64	$${\color{green}-0}$$	$${\color{green}368.7}$$ < $${369.68}$$	$${\color{red}+0}$$	$${\color{green}98.68}$$ < $${99.55}$$
✅	agent_msi	$${\color{green}0}$$	$${\color{green}986.26}$$ < $${987.01}$$	$${\color{red}+0.02}$$	$${\color{green}149.72}$$ < $${150.72}$$
✅	agent_rpm_amd64	$${\color{green}-0}$$	$${\color{green}747.04}$$ < $${752.98}$$	$${\color{red}+0}$$	$${\color{green}186.58}$$ < $${190.03}$$
✅	agent_rpm_amd64_fips	$${\color{green}0}$$	$${\color{green}745.42}$$ < $${751.35}$$	$${\color{red}+0.01}$$	$${\color{green}186.25}$$ < $${189.81}$$
✅	agent_rpm_arm64	$${\color{red}+0}$$	$${\color{green}733.77}$$ < $${739.42}$$	$${\color{red}+0}$$	$${\color{green}168.64}$$ < $${171.23}$$
✅	agent_rpm_arm64_fips	$${\color{green}-0}$$	$${\color{green}732.27}$$ < $${737.91}$$	$${\color{green}-0.01}$$	$${\color{green}167.68}$$ < $${170.22}$$
✅	agent_suse_amd64	$${\color{green}-0}$$	$${\color{green}747.04}$$ < $${752.98}$$	$${\color{red}+0}$$	$${\color{green}186.58}$$ < $${190.03}$$
✅	agent_suse_amd64_fips	$${\color{green}0}$$	$${\color{green}745.42}$$ < $${751.35}$$	$${\color{red}+0.01}$$	$${\color{green}186.25}$$ < $${189.81}$$
✅	agent_suse_arm64	$${\color{red}+0}$$	$${\color{green}733.77}$$ < $${739.42}$$	$${\color{red}+0}$$	$${\color{green}168.64}$$ < $${171.23}$$
✅	agent_suse_arm64_fips	$${\color{green}-0}$$	$${\color{green}732.27}$$ < $${737.91}$$	$${\color{green}-0.01}$$	$${\color{green}167.68}$$ < $${170.22}$$
✅	docker_agent_amd64	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_arm64	$${\color{red}+0}$$	$${\color{green}841.02}$$ < $${858.97}$$	$${\color{green}-0}$$	$${\color{green}268.28}$$ < $${274.36}$$
✅	docker_agent_jmx_amd64	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_jmx_arm64	$${\color{red}+0}$$	$${\color{green}841.02}$$ < $${858.97}$$	$${\color{green}-0}$$	$${\color{green}268.28}$$ < $${274.36}$$
✅	docker_agent_windows1809	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_windows1809_core	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_windows1809_core_jmx	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_windows1809_jmx	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_windows2022	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_windows2022_core	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_windows2022_core_jmx	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_agent_windows2022_jmx	$${\color{green}-0}$$	$${\color{green}830.84}$$ < $${849.39}$$	$${\color{green}-0}$$	$${\color{green}282.17}$$ < $${288.34}$$
✅	docker_cluster_agent_amd64	$${\color{green}0}$$	$${\color{green}259.08}$$ < $${259.73}$$	$${\color{green}-0}$$	$${\color{green}102.83}$$ < $${103.68}$$
✅	docker_cluster_agent_arm64	$${\color{green}0}$$	$${\color{green}273.49}$$ < $${274.24}$$	$${\color{green}-0}$$	$${\color{green}97.55}$$ < $${98.45}$$
✅	docker_cws_instrumentation_amd64	$${\color{red}+0}$$	$${\color{green}7.08}$$ < $${7.12}$$	$${\color{red}+0}$$	$${\color{green}2.95}$$ < $${3.29}$$
✅	docker_cws_instrumentation_arm64	$${\color{green}0}$$	$${\color{green}6.69}$$ < $${6.92}$$	$${\color{green}-0}$$	$${\color{green}2.7}$$ < $${3.07}$$
✅	docker_dogstatsd_amd64	$${\color{red}+0}$$	$${\color{green}38.9}$$ < $${39.57}$$	$${\color{red}+0}$$	$${\color{green}14.94}$$ < $${15.76}$$
✅	docker_dogstatsd_arm64	$${\color{green}0}$$	$${\color{green}37.52}$$ < $${38.2}$$	$${\color{green}-0}$$	$${\color{green}13.96}$$ < $${14.83}$$
✅	dogstatsd_deb_amd64	$${\color{green}0}$$	$${\color{green}30.58}$$ < $${31.52}$$	$${\color{green}-0}$$	$${\color{green}8.03}$$ < $${8.97}$$
✅	dogstatsd_deb_arm64	$${\color{green}0}$$	$${\color{green}29.14}$$ < $${30.08}$$	$${\color{red}+0}$$	$${\color{green}6.97}$$ < $${7.92}$$
✅	dogstatsd_rpm_amd64	$${\color{green}0}$$	$${\color{green}30.58}$$ < $${31.52}$$	$${\color{green}-0}$$	$${\color{green}8.03}$$ < $${8.98}$$
✅	dogstatsd_suse_amd64	$${\color{green}0}$$	$${\color{green}30.58}$$ < $${31.52}$$	$${\color{green}-0}$$	$${\color{green}8.03}$$ < $${8.98}$$
✅	iot_agent_deb_amd64	$${\color{green}0}$$	$${\color{green}59.89}$$ < $${60.17}$$	$${\color{red}+0}$$	$${\color{green}14.97}$$ < $${15.82}$$
✅	iot_agent_deb_arm64	$${\color{green}0}$$	$${\color{green}56.62}$$ < $${56.94}$$	$${\color{red}+0}$$	$${\color{green}12.85}$$ < $${13.86}$$
✅	iot_agent_deb_armhf	$${\color{green}0}$$	$${\color{green}56.07}$$ < $${56.41}$$	$${\color{green}-0.02}$$	$${\color{green}12.98}$$ < $${13.86}$$
✅	iot_agent_rpm_amd64	$${\color{green}0}$$	$${\color{green}59.89}$$ < $${60.18}$$	$${\color{red}+0}$$	$${\color{green}15.0}$$ < $${15.84}$$
✅	iot_agent_rpm_arm64	$${\color{green}0}$$	$${\color{green}56.62}$$ < $${56.94}$$	$${\color{green}-0}$$	$${\color{green}12.87}$$ < $${13.76}$$
✅	iot_agent_suse_amd64	$${\color{green}0}$$	$${\color{green}59.89}$$ < $${60.18}$$	$${\color{red}+0}$$	$${\color{green}15.0}$$ < $${15.84}$$

[grantseltzer]

Can you provide an example of where this is more useful or convenient? We had all ID based references before, and the general feeling was that this connected graph would be easier to use — it means you don’t need to go back and consult a data structure for every lookup

So would all the sub-types (i.e. type of a struct field) be embedded within the struct type definition? Or would it also have its own entry in the type map? So if the subtype is a string, the string type would have to be replicated in all the various places that it's used?

For the sake of decoding I like the idea of having everything be ID based because there's just one rule for resolving type IDs: resolve it in the top level map.

Take a look at the example below for a pointer to a struct with 2 uint32 fields.

1. It's overly verbose and is redundant. In this case uint32 is defined in the Type field for both struct fields, and in the top level map.
2. It allows for more variance in how the IR is used, e.g. do we get the type definition from StructureType.Fields[i].Type.TypeCommon.ID and then go back to the top level Types map, or get all the info from StuctureType.Fields[i].Type... itself??
3. It means that IR generation may change how it generates (i.e. deep or flat) without breaking any APIs but decoding would need to rely on predictable population of the structure.

	irData := ir.Program{
		ID: 1,
		Types: map[ir.TypeID]ir.Type{
			0: &ir.PointerType{
				TypeCommon: ir.TypeCommon{
					ID:       0,
					Name:     "*main.Param",
					ByteSize: 8,
				},
				GoTypeAttributes: ir.GoTypeAttributes{
					GoKind: reflect.Pointer,
				},
				Pointee: &ir.StructureType{
					TypeCommon: ir.TypeCommon{
						ID: 1,
					},
				},
			},
			1: &ir.StructureType{
				TypeCommon: ir.TypeCommon{
					ID:       1,
					Name:     "main.Param",
					ByteSize: 8,
				},
				GoTypeAttributes: ir.GoTypeAttributes{
					GoKind: reflect.Struct,
				},
				Fields: []ir.Field{
					{
						Name:   "idx",
						Offset: 0,
						Type: &ir.BaseType{
							TypeCommon: ir.TypeCommon{
								ID:       2,
								Name:     "uint32",
								ByteSize: 4,
							},
							GoTypeAttributes: ir.GoTypeAttributes{
								GoKind: reflect.Uint32,
							},
						},
					},
					{
						Name:   "random",
						Offset: 4,
						Type: &ir.BaseType{
							TypeCommon: ir.TypeCommon{
								ID:       3,
								Name:     "uint32",
								ByteSize: 4,
							},
							GoTypeAttributes: ir.GoTypeAttributes{
								GoKind: reflect.Uint32,
							},
						},
					},
				},
			},
			2: &ir.BaseType{
				TypeCommon: ir.TypeCommon{
					ID:       2,
					Name:     "uint32",
					ByteSize: 4,
				},
				GoTypeAttributes: ir.GoTypeAttributes{
					GoKind: reflect.Uint32,
				},
			},
			3: &ir.EventRootType{
				TypeCommon: ir.TypeCommon{
					ID:       3,
					Name:     "param",
					ByteSize: 8,
				},
				Expressions: []*ir.RootExpression{
					{
						Offset: 0,
						Expression: ir.Expression{
							Type: &ir.PointerType{
								TypeCommon: ir.TypeCommon{
									Name:     "*main.Param",
									ID:       0,
									ByteSize: 8,
								},
							},
						},
					},
				},
			},
		},
	}

[grantseltzer]

Just found a worse example, look at how many times StringData has to be defined (and how it is currently being generated):

Types: map[ir.TypeID]ir.Type{
			0x1: &ir.GoStringHeaderType{
				StructureType: &ir.StructureType{
					TypeCommon:       ir.TypeCommon{ID: 0x1, Name: "string", ByteSize: 0x10},
					GoTypeAttributes: ir.GoTypeAttributes{GoRuntimeType: 0x45400, GoKind: 0x18},
					Fields: []ir.Field{
						ir.Field{
							Name:   "str",
							Offset: 0x0,
							Type: &ir.PointerType{
								TypeCommon:       ir.TypeCommon{ID: 0x6, Name: "*string.str", ByteSize: 0x8},
								GoTypeAttributes: ir.GoTypeAttributes{},
								Pointee: &ir.GoStringDataType{
									TypeCommon: ir.TypeCommon{ID: 0x5, Name: "string.str", ByteSize: 0x0},
								},
							},
						},
						ir.Field{
							Name:   "len",
							Offset: 0x8,
							Type: &ir.BaseType{
								TypeCommon:       ir.TypeCommon{ID: 0x4, Name: "int", ByteSize: 0x8},
								GoTypeAttributes: ir.GoTypeAttributes{GoRuntimeType: 0x45640, GoKind: 0x2},
							},
						},
					},
				},
				Data: &ir.GoStringDataType{
					TypeCommon: ir.TypeCommon{ID: 0x5, Name: "string.str", ByteSize: 0x0},
				},
			},
			0x2: &ir.PointerType{
				TypeCommon:       ir.TypeCommon{ID: 0x2, Name: "*uint8", ByteSize: 0x8},
				GoTypeAttributes: ir.GoTypeAttributes{GoRuntimeType: 0x2e100, GoKind: 0x16},
				Pointee: &ir.BaseType{
					TypeCommon:       ir.TypeCommon{ID: 0x3, Name: "uint8", ByteSize: 0x1},
					GoTypeAttributes: ir.GoTypeAttributes{GoRuntimeType: 0x45480, GoKind: 0x8},
				},
			},
			0x3: &ir.BaseType{
				TypeCommon:       ir.TypeCommon{ID: 0x3, Name: "uint8", ByteSize: 0x1},
				GoTypeAttributes: ir.GoTypeAttributes{GoRuntimeType: 0x45480, GoKind: 0x8},
			},
			0x4: &ir.BaseType{
				TypeCommon:       ir.TypeCommon{ID: 0x4, Name: "int", ByteSize: 0x8},
				GoTypeAttributes: ir.GoTypeAttributes{GoRuntimeType: 0x45640, GoKind: 0x2},
			},
			0x5: &ir.GoStringDataType{
				TypeCommon: ir.TypeCommon{ID: 0x5, Name: "string.str", ByteSize: 0x0},
			},
			```

[piob-io]

Note that these type structures do not have to be repeated. These type fields are pointers behind the interface. The instances that are pointed too may be, and will be, reused. You can see how it is done here (directly, not via irgen): https://github.com/DataDog/datadog-agent/blob/241a2339b88ce4439dcb3e3d36c5b649cbdc209f/pkg/dyninst/compiler/compile_test.go (file from a pending commit). In particular there will only be single instance of string type implementation.

In a more meta sense, the pointers are like ids, but with extra benefit that they point directly to memory, and don't need an extra id map. I don't view these pointers as redundancy with the id map, because both reference same instances of type implementations. However they are found, they are the same.

I am not sure I follow the third point. The structure will be predictable, but I don't see how it could be broken if the IR gen decides not to necessarily reuse the instances of the type implementation objects (as long as it is still correct data within them).

And the reason for using the pointers, not IDs, is that you don't have to pass around a "type resolver" map around across all the functions that need to process these type instances. Not a huge deal to do that, but also a nice convenience not to have to.

[ajwerner]

To clarify, are you saying that the primary motivation for this change is to make it easier to construct ir by hand for testing? Is there any motivation that's not related to testing ergonomics?

[ajwerner]

I will say that there are ways in which using an ID-oriented approach can be more efficient from an object-size perspective and a GC overhead perspective, but we're not in a place of optimizing at that level just yet.

[ajwerner]

I think another thing that maybe wasn't said but should have been said is there should be an explicit assumption that for a given instance of a program there's exactly one instance of a Type with a given ID and that you can check that with pointer equality.