Data flow: Store/load matching in pruning stage 3

[hvitved]

Example tuple counts

C#: HardCodedCredentials.ql on mono

Before

#	n	stage	nodes	fields	conscand	states	tuples	calledges	tfnodes	tftuples
1	10	1 Fwd	3,498,701	41,634	-1	1	4,985,909	-1	-1	-1
2	15	1 Rev	1,988,816	32,443	-1	1	3,015,751	1,927,923	-1	-1
3	20	2 Fwd	1,227,365	14,291	17,410	1	2,377,149	999,216	0	0
4	25	2 Rev	816,624	9,731	11,168	1	1,155,022	368,049	0	0
5	30	3 Fwd	445,018	6,897	26,750	1	3,140,440	293,820	44,062	471,082
6	35	3 Rev	984	45	55	1	1,231	1,194	617	634
7	40	4 Fwd	95	4	5	1	105	30	0	0
8	45	4 Rev	95	4	5	1	99	30	0	0
9	50	5 Fwd	95	4	5	1	143	30	0	0
10	55	5 Rev	95	4	5	1	129	30	0	0
11	60	6 Fwd	95	4	5	1	136	30	0	0
12	65	6 Rev	95	4	5	1	129	30	0	0

After

#	n	stage	nodes	fields	conscand	states	tuples	calledges	tfnodes	tftuples
1	10	1 Fwd	3,498,701	41,634	-1	1	4,985,909	-1	-1	-1
2	15	1 Rev	1,988,816	32,443	-1	1	3,015,751	1,927,923	-1	-1
3	20	2 Fwd	1,227,365	14,291	17,410	1	2,377,149	999,216	0	0
4	25	2 Rev	816,624	9,731	11,168	1	1,155,022	368,049	0	0
5	30	3 Fwd	365,688	4,669	14,873	1	1,781,588	246,743	40,080	312,192
6	35	3 Rev	507	17	24	1	609	168	491	500
7	40	4 Fwd	95	4	5	1	105	30	0	0
8	45	4 Rev	95	4	5	1	99	30	0	0
9	50	5 Fwd	95	4	5	1	143	30	0	0
10	55	5 Rev	95	4	5	1	129	30	0	0
11	60	6 Fwd	95	4	5	1	136	30	0	0
12	65	6 Rev	95	4	5	1	129	30	0	0

Javascript: InsecureRandomness.ql on microsoft_vscode

Before

#	n	stage	nodes	fields	conscand	states	tuples	calledges	tfnodes	tftuples
1	10	1 Fwd	862,069	11,800	-1	1	1,116,618	-1	-1	-1
2	15	1 Rev	351,604	9,837	-1	1	476,835	77,147	-1	-1
3	20	2 Fwd	132,188	3,784	5,140	1	278,795	35,111	0	0
4	25	2 Rev	96,795	3,001	4,099	1	158,086	25,646	0	0
5	30	3 Fwd	86,505	2,789	5,741	1	319,508	24,224	84	1,179
6	35	3 Rev	80,912	2,649	3,987	1	164,567	22,598	76	1,024
7	40	4 Fwd	68,557	1,808	147,485	1	147,992,833	20,336	74	1,108
8	45	4 Rev	52,732	1,472	96,188	1	27,297,733	15,697	66	904
9	50	5 Fwd	2,218	37	65	1	3,365	962	4	12
10	55	5 Rev	96	2	2	1	96	61	0	0
11	60	6 Fwd	96	2	3	1	116	-1	-1	-1
12	65	6 Rev	84	2	3	1	84	-1	-1	-1

After

#	n	stage	nodes	fields	conscand	states	tuples	calledges	tfnodes	tftuples
1	10	1 Fwd	862,069	11,800	-1	1	1,116,618	-1	-1	-1
2	15	1 Rev	351,604	9,837	-1	1	476,835	77,147	-1	-1
3	20	2 Fwd	132,188	3,784	5,140	1	278,795	35,111	0	0
4	25	2 Rev	96,795	3,001	4,099	1	158,086	25,646	0	0
5	30	3 Fwd	21,495	373	568	1	42,953	6,513	48	116
6	35	3 Rev	7,315	197	259	1	9,942	2,426	26	64
7	40	4 Fwd	4,475	114	366	1	11,854	1,469	12	32
8	45	4 Rev	4,026	106	190	1	6,632	1,297	12	32
9	50	5 Fwd	351	16	18	1	421	120	0	0
10	55	5 Rev	96	2	2	1	96	61	0	0
11	60	6 Fwd	96	2	3	1	116	-1	-1	-1
12	65	6 Rev	84	2	3	1	84	-1	-1	-1

[asgerf]

This looks very promising for the JS data flow migration. 🎉 It fixes the four largest regressions we had so far. See DCA run showing the effects of cherry-picking this PR on top of the JS data flow migration branch.