Io handled by ethash

[LefterisJP]

• Work in progress
• For clarity and for future extendability have 2 different source files that are conditionally compiled depending on being in Windows or not for the function that makes calls to OS dependent functions.
• I created these 2 functions that are intended to be used by the clients by looking at how the C++ client writes the DAG in the disk.
  • One would call ethash_io_prepare() first to make sure the directory exists and check if the memo file's contents match with the current block hash.
  • If there is a mismatch then you would call ethash_io_write() which would compute the full data just like ethash_prep_full() would, write it to the disk along with the memo file and also return them.

Have not yet tried to use it from the C++ client but need to first see if the PR will build in Windows.

[gavofyork]

• take seedhash into the creation rather than block number (which contains additional useless information).
• this is all fine, but ultimately it should be a completely opaque interface and look after memory management - at present we have:

void ethash_prep_full(void *full, ethash_params const *params, void const *cache);
void ethash_compute_full(ethash_return_value *ret, void const *full, ethash_params const *params, const uint8_t header_hash[32], const uint64_t nonce);

these should become:

typedef void(*Callback)(unsigned);
void* ethash_new_full(ethash_params const* params, void const* cache, const uint8_t seed[32]), CallBack c);
void* ethash_compute_full(ethash_return_value *ret, void const *full, ethash_params const *params, const uint8_t header_hash[32], const uint64_t nonce);
void ethash_delete_full(void const* full);

ethash_new_full should handle caching opaquely and call the Callback 100 times passing 0..99 as the argument.

Before, it was up to the caller to allocate the memory required, passing it in as a void*. Now it should be allocated by ethash_new_full and deallocated by ethash_delete_full.

ethash_compute_full should act as it did before.

May be worth splitting it up into two PRs - integrating memory management separately.

[LefterisJP]

I typedefed uint8_t[32] as a blockhash_t. For now it's only in io.h but this should go all over the codebase. It's much more readable and we also get free typechecking this way.

[gavofyork]

new API spec:

typedef void(*Callback)(unsigned);
typedef void const* ethash_full_t;
ethash_full_t ethash_new_full(ethash_params const* params, void const* cache, const uint8_t seed[32]), CallBack c);
void ethash_compute_full(ethash_return_value *ret, ethash_full_t full, ethash_params const *params, const uint8_t header_hash[32], const uint64_t nonce);
void ethash_delete_full(ethash_full_t full);

[gavofyork]

spaces or tabs, not both :)

[gavofyork]

newer API spec:

typedef int(*Callback)(unsigned);
typedef void const* ethash_full_t;
ethash_full_t ethash_new_full(ethash_params const* params, void const* cache, const uint8_t seed[32]), CallBack c);
void ethash_compute_full(ethash_return_value *ret, ethash_full_t full, ethash_params const *params, const uint8_t header_hash[32], const uint64_t nonce);
void ethash_delete_full(ethash_full_t full);

non-zero return from Callback means "cancel DAG creation" - this should cause an immediate return of ethash_new_full with 0.

an object of type ethash_full_t may be tested for validity with != 0

Example usage:

int callback(unsigned _progress)
{
  printf("\rGenerating DAG. %d%% done...", _progress);
  return 0;
}
void main()
{
  ethash_params p;
  uint8_t const* cache;
  uint8_t seed[32];
  // TODO: populate p, seed, cache
  ethash_full_t dag = ethash_new_full(&p, cache, seed, &callback);
  if (!dag)
  {
    printf("Failed generating DAG :-(\n");
    exit(-1);
  }
  printf("DAG Generated OK!\n");

  uint8_t headerHash[32];
  // TODO: populate headerHash
  uint64_t nonce = time(0);
  ethash_return_value ret;
  for (; !isWinner(ret); nonce++)
    ethash_compute_full(&ret, dag, &p, headerHash, nonce);
  printf("Got winner! nonce is %d\n", nonce);
  ethash_delete_full(dag);
}

[debris]

better way:

find_package(Boost 1.54.0 REQUIRED COMPONENTS unit_test_framework system filesystem)

[LefterisJP]

this should be BOOST 1.48.0 since this is only for testing and C++11 and is not the same as the rest of the codebase. Basically boost 1.48.0 is the max we can get in travis-CI so this is kind of a limitation :(

[LefterisJP]

rebased