[ALL] Start work on restructuring based on hashes (will not compile)

This set of changes is the start of turning over to a Hash based system instead of a Rom/Dat based system. It's a long process but it will be worth it in the end.
This commit is contained in:
Matt Nadareski
2016-08-29 16:33:07 -07:00
parent a5dae7e693
commit 0b97b046c8
18 changed files with 2094 additions and 226 deletions

View File

@@ -21,7 +21,7 @@ namespace SabreTools.Helper
public static Rom GetSingleFileInfo(string input, bool noMD5 = false, bool noSHA1 = false, long offset = 0)
{
// Add safeguard if file doesn't exist
if (!System.IO.File.Exists(input))
if (!File.Exists(input))
{
return new Rom();
}
@@ -44,7 +44,7 @@ namespace SabreTools.Helper
using (Crc32 crc = new Crc32())
using (MD5 md5 = MD5.Create())
using (SHA1 sha1 = SHA1.Create())
using (FileStream fs = System.IO.File.OpenRead(input))
using (FileStream fs = File.OpenRead(input))
{
// Seek to the starting position, if one is set
if (offset > 0)
@@ -186,6 +186,104 @@ namespace SabreTools.Helper
return outroms;
}
/// <summary>
/// Merge an arbitrary set of ROMs based on the supplied information
/// </summary>
/// <param name="inhashes">List of HashData objects representing the roms to be merged</param>
/// <param name="logger">Logger object for console and/or file output</param>
/// <returns>A List of HashData objects representing the merged roms</returns>
public static List<HashData> Merge(List<HashData> inhashes, Logger logger)
{
// Create output list
List<HashData> outroms = new List<HashData>();
// Check for null or blank roms first
if (inhashes == null || inhashes.Count == 0)
{
return outroms;
}
// Then deduplicate them by checking to see if data matches previous saved roms
foreach (HashData hash in inhashes)
{
// If it's a nodump, add and skip
if (hash.Roms[0].Nodump)
{
outroms.Add(hash);
continue;
}
// If it's the first rom in the list, don't touch it
if (outroms.Count != 0)
{
// Check if the rom is a duplicate
DupeType dupetype = DupeType.None;
HashData savedHash = new HashData();
int pos = -1;
for (int i = 0; i < outroms.Count; i++)
{
HashData lastrom = outroms[i];
RomData savedRom = savedHash.Roms[0];
MachineData savedMachine = savedRom.Machine;
// Get the duplicate status
dupetype = GetDuplicateStatus(hash, lastrom, logger);
// If it's a duplicate, skip adding it to the output but add any missing information
if (dupetype != DupeType.None)
{
savedHash = lastrom;
pos = i;
savedHash.CRC = (savedHash.CRC == null && hash.CRC != null ? hash.CRC : savedHash.CRC);
savedHash.MD5 = (savedHash.MD5 == null && hash.MD5 != null ? hash.MD5 : savedHash.MD5);
savedHash.SHA1 = (savedHash.SHA1 == null && hash.SHA1 != null ? hash.SHA1 : savedHash.SHA1);
savedRom.DupeType = dupetype;
// If the current system has a lower ID than the previous, set the system accordingly
if (hash.Roms[0].Machine.SystemID < savedMachine.SystemID)
{
savedMachine.SystemID = hash.Roms[0].Machine.SystemID;
savedMachine.System = hash.Roms[0].Machine.System;
savedMachine.Name = hash.Roms[0].Machine.Name;
savedRom.Name = hash.Roms[0].Name;
}
// If the current source has a lower ID than the previous, set the source accordingly
if (hash.Roms[0].Machine.SourceID < savedMachine.SourceID)
{
savedMachine.SourceID = hash.Roms[0].Machine.SourceID;
savedMachine.Source = hash.Roms[0].Machine.Source;
savedMachine.Name = hash.Roms[0].Machine.Name;
savedRom.Name = hash.Roms[0].Name;
}
break;
}
}
// If no duplicate is found, add it to the list
if (dupetype == DupeType.None)
{
outroms.Add(hash);
}
// Otherwise, if a new rom information is found, add that
else
{
outroms.RemoveAt(pos);
outroms.Insert(pos, savedHash);
}
}
else
{
outroms.Add(hash);
}
}
// Then return the result
return outroms;
}
/// <summary>
/// List all duplicates found in a DAT based on a rom
/// </summary>
@@ -348,5 +446,51 @@ namespace SabreTools.Helper
return hash;
}
/// <summary>
/// Clean a hash byte array and pad to the correct size
/// </summary>
/// <param name="hash">Hash byte array to sanitize</param>
/// <param name="padding">Amount of bytes to pad to</param>
/// <returns>Cleaned byte array</returns>
public static byte[] CleanHashData(byte[] hash, int padding)
{
// If we have a null hash or a <=0 padding, return the hash
if (hash == null || padding <= 0)
{
return hash;
}
// If we have a hash longer than the padding, trim and return
if (hash.Length > padding)
{
return hash.Take(padding).ToArray();
}
// If we have a hash of the correct length, return
if (hash.Length == padding)
{
return hash;
}
// Otherwise get the output byte array of the correct length
byte[] newhash = new byte[padding];
// Then write the proper number of empty bytes
int padNeeded = padding - hash.Length;
int index = 0;
for (index = 0; index < padNeeded; index++)
{
newhash[index] = 0x00;
}
// Now add the original hash
for (int i = 0; i < hash.Length; i++)
{
newhash[index + i] = hash[index];
}
return newhash;
}
}
}