effekt Add resource finalization to handlers

This PR aims to add resource finalization to handlers. This was already implemented under the branch feature/finalizer, however, the branch is now hilariously outdated and solving the merge conflicts almost impossible. Thus, I manually try to re-implement it.

Nov 19 '24 15:11 dvdvgt

The changes to the inliner seem to break some tests. For example, consider the JS output for the file anf.effekt.md:

Without inlining

function traverse_0(e_0, Fresh_2, Bind_0, ks_331, k_263) {
  function b_k_38(n_6, ks_332, k_264) {
    return () => k_264(new CRet_0(new CLit_0(n_6)), ks_332);
  }
  function b_k_39(n_7, ks_333, k_265) {
    return () => k_265(new CRet_0(new CVar_0(n_7)), ks_333);
  }
  function b_k_40(name_7, arg_3, ks_334, k_266) {
    return () =>
      traverse_0(arg_3, Fresh_2, Bind_0, ks_334, (v_r_54, ks_335) =>
        Bind_0.Bind_1(v_r_54, ks_335, (v_r_55, ks_336) =>
          () => k_266(new CApp_0(name_7, v_r_55), ks_336)));
  }
  function b_k_41(x_22, b_0, body_3, ks_337, k_267) {
    return () =>
      bindHere_0((Bind_2, ks_338, k_268) =>
        () => traverse_0(b_0, Fresh_2, Bind_2, ks_338, k_268), Fresh_2, ks_337, (v_r_56, ks_339) =>
        () =>
          bindHere_0((Bind_3, ks_340, k_269) =>
            () => traverse_0(body_3, Fresh_2, Bind_3, ks_340, k_269), Fresh_2, ks_339, (v_r_57, ks_341) =>
            () => k_267(new CLet_0(x_22, v_r_56, v_r_57), ks_341)));
  }
  switch (e_0.__tag) {
    case 0: 
      const v_y_18 = e_0.value_1;
      return () => b_k_38(v_y_18, ks_331, k_263);
    case 1: 
      const v_y_19 = e_0.name_0;
      return () => b_k_39(v_y_19, ks_331, k_263);
    case 3: 
      const v_y_20 = e_0.name_2;
      const v_y_21 = e_0.arg_0;
      return () => b_k_40(v_y_20, v_y_21, ks_331, k_263);
    case 2: 
      const v_y_22 = e_0.name_1;
      const v_y_23 = e_0.binding_0;
      const v_y_24 = e_0.body_0;
      return () => b_k_41(v_y_22, v_y_23, v_y_24, ks_331, k_263);
  }
}

with inlining

function traverse_0(e_4, Fresh_0, Bind_0, ks_2182, k_1778) {
  switch (e_4.__tag) {
    case 0: 
      const v_y_356 = e_4.value_11;
      return () => k_1778(new CRet_0(new CLit_0(v_y_356)), ks_2182);
    case 1: 
      const v_y_357 = e_4.name_7;
      return () => k_1778(new CRet_0(new CVar_0(v_y_357)), ks_2182);
    case 3: 
      const v_y_358 = e_4.name_9;
      const v_y_359 = e_4.arg_0;
      return () =>
        traverse_0(v_y_359, Fresh_0, Bind_0, ks_2182, (v_r_1297, ks_2183) =>
          Bind_0.Bind_1(v_r_1297, ks_2183, (v_r_1298, ks_2184) =>
            () => k_1778(new CApp_0(v_y_358, v_r_1298), ks_2184)));
    case 2: 
      const v_y_360 = e_4.binding_0;
      return RESET((p_174, ks_2185, k_1779) =>
        () =>
          traverse_0(v_y_360, Fresh_0, {
            Bind_1: (e_5, ks_2186, k_1780) =>
              SHIFT(p_174, (k_1781, ks_2187, k_1782) =>
                Fresh_0.Fresh_1(ks_2187, (id_2, ks_2188) =>
                  RESUME(k_1781, (ks_2189, k_1783) =>
                    k_1783(new CVar_0(id_2), ks_2189), false, ks_2188, (v_r_1299, ks_2190) =>
                    k_1782(new CLet_0(id_2, e_5, v_r_1299), ks_2190))), ks_2186, k_1780, null)
          }, ks_2185, k_1779), null, null, null, ks_2182, (v_r_1300, ks_2191) =>
        RESET((p_175, ks_2192, k_1784) =>
          () =>
            traverse_0(v_y_361, Fresh_0, {
              Bind_1: (e_6, ks_2193, k_1785) =>
                SHIFT(p_175, (k_1786, ks_2194, k_1787) =>
                  Fresh_0.Fresh_1(ks_2194, (id_3, ks_2195) =>
                    RESUME(k_1786, (ks_2196, k_1788) =>
                      k_1788(new CVar_0(id_3), ks_2196), false, ks_2195, (v_r_1301, ks_2197) =>
                      k_1787(new CLet_0(id_3, e_6, v_r_1301), ks_2197))), ks_2193, k_1785, null)
            }, ks_2192, k_1784), null, null, null, ks_2191, (v_r_1302, ks_2198) =>
          () => k_1778(new CLet_0(v_y_362, v_r_1300, v_r_1302), ks_2198)));
  }
}

We can see that two of the pattern matching values are missing (inlined?), however, their names still appear unbound (v_y_361). Thus, two of the tests fail because of this.

Dec 02 '24 16:12 dvdvgt

This issue ↑ seems similar to the problem detected in https://github.com/effekt-lang/effekt/pull/561#issuecomment-2498899363

Dec 02 '24 16:12 jiribenes

Sadly, the fix proposed in https://github.com/effekt-lang/effekt/pull/561#issuecomment-2513199799 does not seem to fix the failing tests for anf.effekt.md and prettyprinter.effekt.md. I am still not quite sure why.

Dec 03 '24 14:12 dvdvgt

The changes to the inliner seem to break some tests.

This was actually due to a faulty free variable function which caused some definitions used in continuations not be marked as used and consequently deleted as dead code. This was a fun debugging session 🙃

Dec 20 '24 22:12 dvdvgt