functions - Defining Tags in Reap

I am trying to use Sow/Reap to replace Append in my code. The problem is that I need to define the tags within my code, but doing so doesn't reap the desired results. A simplified version of my code is

Clear[tag];
Reap[tag = {a,b}; Sow[1, a];Sow[2,b];, tag]

However, if I define my tag outside of Reap, then it works

Clear[tag];
tag = {a,b};
Reap[Sow[1, a];Sow[2,b];, tag]

Any ideas as to how I can define tags within my code?

Answer

Reap does not work that way: when you provide a list of tags for Reap only these tags are "watched for" and collected. This allows for much better memory management than collecting for all tags and discarding at the end.

In essence it works something like this for declared tags 1, 2, 3:

{one, two, three} = {{},{},{}};

sow[x_, 1] := (one   = {one, x};   x)
sow[x_, 2] := (two   = {two, x};   x)
sow[x_, 3] := (three = {three, x}; x)
sow[x_, _] := x


SeedRandom[1]
Do[sow[i, RandomInteger[10^6]], {i, 10^7}]

Flatten /@ {one, two, three}

MaxMemoryUsed[]

{{30003, 1238414, 1529333, 3074569, 3401105, 4162839, 4715096, 5855206, 5971795, 6984287},

 {238730, 652982, 946353, 1821955, 2018277, 2065726, 4483637,
  4591412, 4733380, 4920935, 5283043, 5816356, 8272260, 8409277},
 {2793919, 2803799, 3784289, 5018439, 6380588, 8799862, 9301537}}

15034792

You can see that advance knowledge of the tags is needed to set this up. Compare to the memory requirements of this:

SeedRandom[1]

Cases[

 {#[[All, 1]], #[[1, 2]]} & /@
   Table[{i, RandomInteger[10^6]}, {i, 10^7}] ~GatherBy~ Last,
 {_, 1 | 2 | 3},
 {1}
]

MaxMemoryUsed[]

{{{30003, 1238414, 1529333, 3074569, 3401105, 4162839, 4715096, 

   5855206, 5971795, 6984287}, 1},
 {{238730, 652982, 946353, 1821955, 2018277, 2065726, 4483637, 
   4591412, 4733380, 4920935, 5283043, 5816356, 8272260, 8409277}, 2},
 {{2793919, 2803799, 3784289, 5018439, 6380588, 8799862, 9301537}, 3}}

605960144

---

If you wish to replace Append with something more efficient, consider using linked lists as I did in the sow example above, or the Internal`Bag class of functions.