1
0
mirror of https://github.com/fadden/6502bench.git synced 2024-11-04 15:05:03 +00:00
6502bench/Asm65/Formatter.cs
Andy McFadden 86ead987d5 Tweak OMF converter
Changed bank-start comments to notes, added a summary to the top-of-file
comment.

Also, fixed a bug where the app settings dialog wasn't identifying
display settings as a preset for 64tass and cc65.
2020-06-30 11:48:11 -07:00

1130 lines
48 KiB
C#

/*
* Copyright 2018 faddenSoft
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Text;
using AddressMode = Asm65.OpDef.AddressMode;
namespace Asm65 {
/// <summary>
/// Functions used for formatting bits of 65xx code into human-readable form.
///
/// There are a variety of ways to format a given thing, based on personal preference
/// (e.g. whether opcodes are upper- or lower-case) and assembler syntax requirements.
///
/// The functions in this class serve two purposes: (1) produce consistent output
/// throughout the program; (2) cache format strings and other components to reduce
/// string manipulation overhead. Note the caching is per-Formatter, so it's best to
/// create just one and share it around.
///
/// The configuration of a Formatter may not be altered once created. This is important
/// in situations where we compute output size in one pass and generate it in another,
/// because it guarantees that a given Formatter object will produce the same number of
/// lines of output.
///
/// NOTE: if the CpuDef changes, the cached values in the Formatter will become invalid
/// (e.g. mOpcodeStrings). Discard the Formatter and create a new one. (This could be
/// fixed by keying off of the OpDef instead of OpDef.Opcode, but that's less convenient.)
/// </summary>
public class Formatter {
/// <summary>
/// Various format configuration options. Fill one of these out and pass it to
/// the Formatter constructor.
/// </summary>
public struct FormatConfig {
// alpha case for some case-insensitive items
public bool mUpperHexDigits; // display hex values in upper case?
public bool mUpperOpcodes; // display opcodes in upper case?
public bool mUpperPseudoOpcodes; // display pseudo-opcodes in upper case?
public bool mUpperOperandA; // display acc operand in upper case?
public bool mUpperOperandS; // display stack operand in upper case?
public bool mUpperOperandXY; // display index register operand in upper case?
public bool mAddSpaceLongComment; // insert space after delimiter for long comments?
// functional changes to assembly output
public bool mSuppressHexNotation; // omit '$' before hex digits
public bool mSuppressImpliedAcc; // emit just "LSR" rather than "LSR A"?
public bool mBankSelectBackQuote; // use '`' rather than '^' for bank selector?
public string mForceDirectOperandPrefix; // these may be null or empty
public string mForceAbsOpcodeSuffix;
public string mForceAbsOperandPrefix;
public string mForceDirectOpcodeSuffix;
public string mForceLongOpcodeSuffix;
public string mForceLongOperandPrefix;
public string mLocalVariableLabelPrefix; // e.g. Merlin 32 puts ']' before var names
public string mNonUniqueLabelPrefix; // e.g. ':' or '@' before local label
public string mEndOfLineCommentDelimiter; // usually ';'
public string mFullLineCommentDelimiterBase; // usually ';' or '*', WITHOUT extra space
public string mBoxLineCommentDelimiter; // usually blank or ';'
// delimiter patterns for single character constants
public DelimiterSet mCharDelimiters;
public DelimiterSet mStringDelimiters;
// miscellaneous
public bool mSpacesBetweenBytes; // "20edfd" vs. "20 ed fd"
public bool mCommaSeparatedDense; // "20edfd" vs. "$20,$ed,$fd"
// hex dumps
public bool mHexDumpAsciiOnly; // disallow non-ASCII chars in hex dumps?
public enum CharConvMode {
// TODO(maybe): just pass in a CharEncoding.Convert delegate
Unknown = 0,
Ascii,
LowHighAscii,
C64Petscii,
C64ScreenCode
};
public CharConvMode mHexDumpCharConvMode; // character conversion mode for dumps
// This determines what operators are available and what their precedence is.
// Hopefully we don't need a separate mode for every assembler in existence.
public enum ExpressionMode { Unknown = 0, Common, Cc65, Merlin };
public ExpressionMode mExpressionMode; // symbol rendering mode
// Deserialization helper.
public static ExpressionMode ParseExpressionMode(string str) {
ExpressionMode em = ExpressionMode.Common;
if (!string.IsNullOrEmpty(str)) {
if (Enum.TryParse<ExpressionMode>(str, out ExpressionMode pem)) {
em = pem;
}
}
return em;
}
// TODO: FormatConfig should be a class with properties so we can avoid this nonsense
public void Normalize() {
if (mForceDirectOperandPrefix == null) {
mForceDirectOperandPrefix = string.Empty;
}
if (mForceAbsOpcodeSuffix == null) {
mForceAbsOpcodeSuffix = string.Empty;
}
if (mForceAbsOperandPrefix == null) {
mForceAbsOperandPrefix = string.Empty;
}
if (mForceDirectOpcodeSuffix == null) {
mForceDirectOpcodeSuffix = string.Empty;
}
if (mForceLongOpcodeSuffix == null) {
mForceLongOpcodeSuffix = string.Empty;
}
if (mForceLongOperandPrefix == null) {
mForceLongOperandPrefix = string.Empty;
}
if (mLocalVariableLabelPrefix == null) {
mLocalVariableLabelPrefix = string.Empty;
}
if (mNonUniqueLabelPrefix == null) {
mNonUniqueLabelPrefix = string.Empty;
}
if (mEndOfLineCommentDelimiter == null) {
mEndOfLineCommentDelimiter = string.Empty;
}
if (mFullLineCommentDelimiterBase == null) {
mFullLineCommentDelimiterBase = string.Empty;
}
if (mBoxLineCommentDelimiter == null) {
mBoxLineCommentDelimiter = string.Empty;
}
}
}
#region Text Delimiters
/// <summary>
/// Container for character and string delimiter pieces. Instances are immutable.
/// </summary>
/// <remarks>
/// For single-character operands, a simple concatenation of the four fields, with the
/// character in the middle, is performed.
///
/// For strings, the prefix is included at the start of the first line, but not included
/// on subsequent lines. This is primarily intended for the on-screen display, not
/// assembly source generation. The suffix is not used at all.
/// </remarks>
public class DelimiterDef {
public string Prefix { get; private set; }
public char OpenDelim { get; private set; }
public char CloseDelim { get; private set; }
public string Suffix { get; private set; }
public string FormatStr { get; private set; }
public DelimiterDef(char delim) : this(string.Empty, delim, delim, string.Empty) {
}
public DelimiterDef(string prefix, char openDelim, char closeDelim, string suffix) {
Debug.Assert(prefix != null);
Debug.Assert(suffix != null);
Prefix = prefix;
OpenDelim = openDelim;
CloseDelim = closeDelim;
Suffix = suffix;
// Generate format string.
StringBuilder sb = new StringBuilder();
sb.Append(Prefix);
sb.Append(OpenDelim);
sb.Append("{0}");
sb.Append(CloseDelim);
sb.Append(Suffix);
FormatStr = sb.ToString();
}
public override string ToString() {
return Prefix + OpenDelim + '#' + CloseDelim + Suffix;
}
}
public static readonly DelimiterDef SINGLE_QUOTE_DELIM = new DelimiterDef('\'');
public static readonly DelimiterDef DOUBLE_QUOTE_DELIM = new DelimiterDef('"');
public class DelimiterSet {
private Dictionary<CharEncoding.Encoding, DelimiterDef> mDelimiters =
new Dictionary<CharEncoding.Encoding, DelimiterDef>();
/// <summary>
/// Returns the specified DelimiterDef, or null if not found.
/// </summary>
public DelimiterDef Get(CharEncoding.Encoding enc) {
mDelimiters.TryGetValue(enc, out DelimiterDef def);
return def;
}
public void Set(CharEncoding.Encoding enc, DelimiterDef def) {
mDelimiters[enc] = def;
}
public override string ToString() {
StringBuilder sb = new StringBuilder();
foreach (KeyValuePair<CharEncoding.Encoding, DelimiterDef> kvp in mDelimiters) {
sb.Append("[" + kvp.Key + ": " + kvp.Value + "]");
}
return sb.ToString();
}
public static DelimiterSet GetDefaultCharDelimiters() {
DelimiterSet chrDel = new DelimiterSet();
chrDel.Set(CharEncoding.Encoding.Ascii,
new DelimiterDef(string.Empty, '\u2018', '\u2019', string.Empty));
chrDel.Set(CharEncoding.Encoding.HighAscii,
new DelimiterDef(string.Empty, '\u2018', '\u2019', " | $80"));
chrDel.Set(CharEncoding.Encoding.C64Petscii,
new DelimiterDef("pet:", '\u2018', '\u2019', string.Empty));
chrDel.Set(CharEncoding.Encoding.C64ScreenCode,
new DelimiterDef("scr:", '\u2018', '\u2019', string.Empty));
return chrDel;
}
public static DelimiterSet GetDefaultStringDelimiters() {
DelimiterSet strDel = new DelimiterSet();
strDel.Set(CharEncoding.Encoding.Ascii,
new DelimiterDef(string.Empty, '\u201c', '\u201d', string.Empty));
strDel.Set(CharEncoding.Encoding.HighAscii,
new DelimiterDef("\u2191", '\u201c', '\u201d', string.Empty));
strDel.Set(CharEncoding.Encoding.C64Petscii,
new DelimiterDef("pet:", '\u201c', '\u201d', string.Empty));
strDel.Set(CharEncoding.Encoding.C64ScreenCode,
new DelimiterDef("scr:", '\u201c', '\u201d', string.Empty));
return strDel;
}
/// <summary>
/// Serializes a DelimiterSet.
/// </summary>
/// <remarks>
/// Can't use Javascript from a .NET Standard library. XmlSerializer doesn't
/// handle Lists or Dictionaries. Do it the old-fashioned way.
/// </remarks>
public string Serialize() {
Debug.Assert(mDelimiters.Count < 10);
StringBuilder sb = new StringBuilder();
sb.Append('*'); // if the format changes, start with something else
foreach (KeyValuePair<CharEncoding.Encoding, DelimiterDef> kvp in mDelimiters) {
string name = kvp.Key.ToString();
AddLenString(sb, name);
AddLenString(sb, kvp.Value.Prefix);
sb.Append(kvp.Value.OpenDelim);
sb.Append(kvp.Value.CloseDelim);
AddLenString(sb, kvp.Value.Suffix);
}
sb.Append('!');
return sb.ToString();
}
private void AddLenString(StringBuilder sb, string str) {
sb.Append(str.Length.ToString());
sb.Append(',');
sb.Append(str);
}
public static DelimiterSet Deserialize(string cereal) {
try {
DelimiterSet delimSet = new DelimiterSet();
int offset = 0;
if (cereal[offset++] != '*') {
throw new Exception("missing leading asterisk");
}
while (cereal[offset] != '!') {
string str = GetLenString(cereal, ref offset);
if (!Enum.TryParse(str, out CharEncoding.Encoding enc)) {
Debug.WriteLine("Ignoring unknown encoding " + str);
enc = CharEncoding.Encoding.Unknown;
}
string prefix = GetLenString(cereal, ref offset);
char open = cereal[offset++];
char close = cereal[offset++];
string suffix = GetLenString(cereal, ref offset);
if (enc != CharEncoding.Encoding.Unknown) {
delimSet.Set(enc, new DelimiterDef(prefix, open, close, suffix));
}
}
return delimSet;
} catch (Exception ex) {
Debug.WriteLine("DelimiterSet deserialization failed: " + ex.Message);
return new DelimiterSet();
}
}
private static string GetLenString(string str, ref int offset) {
int commaIndex = str.IndexOf(',', offset);
if (commaIndex < 0) {
throw new Exception("no comma in length string");
}
string lenStr = str.Substring(offset, commaIndex - offset);
int len = int.Parse(lenStr);
string resultStr = str.Substring(commaIndex + 1, len);
offset = commaIndex + 1 + len;
return resultStr;
}
}
#endregion Text Delimiters
private static readonly char[] sHexCharsLower = {
'0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'a', 'b', 'c', 'd', 'e', 'f'
};
private static readonly char[] sHexCharsUpper = {
'0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'A', 'B', 'C', 'D', 'E', 'F'
};
/// <summary>
/// Formatter configuration options. Fixed at construction time.
/// </summary>
private FormatConfig mFormatConfig;
/// <summary>
/// Get a copy of the format config.
/// </summary>
public FormatConfig Config { get { return mFormatConfig; } }
// Cached bits and pieces.
char mHexFmtChar;
string mHexPrefix;
string mAccChar;
char mXregChar;
char mYregChar;
char mSregChar;
// Format string for offsets.
private string mOffset24Format;
// Format strings for addresses.
private string mAddrFormatNoBank;
private string mAddrFormatWithBank;
// Generated opcode strings. The index is the bitwise OR of the opcode value and
// the disambiguation value. In most cases this just helps us avoid calling
// ToUpper incessantly.
private Dictionary<int, string> mOpcodeStrings = new Dictionary<int, string>();
// Generated pseudo-opcode strings.
private Dictionary<string, string> mPseudoOpStrings = new Dictionary<string, string>();
// Generated format strings for operands. The index is the bitwise OR of the
// address mode and the disambiguation value.
private Dictionary<int, string> mOperandFormats = new Dictionary<int, string>();
// Generated format strings for bytes.
private const int MAX_BYTE_DUMP = 4;
private string[] mByteDumpFormats = new string[MAX_BYTE_DUMP];
// Generated format strings for hex values.
private string[] mHexValueFormats = new string[4];
private string mFullLineCommentDelimiterPlus;
// Buffer to use when generating hex dump lines.
private char[] mHexDumpBuffer;
private CharEncoding.Convert mHexDumpCharConv;
/// <summary>
/// A 16-character array with 0-9a-f, for hex conversions. The letters will be
/// upper or lower case, per the format config.
/// </summary>
public char[] HexDigits {
get {
return mFormatConfig.mUpperHexDigits ? sHexCharsUpper : sHexCharsLower;
}
}
/// <summary>
/// String to put between the operand and the end-of-line comment.
/// </summary>
public string EndOfLineCommentDelimiter {
get { return mFormatConfig.mEndOfLineCommentDelimiter; }
}
/// <summary>
/// String to put at the start of a line with a full-line comment.
/// </summary>
public string FullLineCommentDelimiter {
get { return mFullLineCommentDelimiterPlus; }
}
/// <summary>
/// String to put at the start of a line that has a box comment. This is usually
/// blank, as it's only needed if the assembler doesn't recognize the box character
/// as a comment.
/// </summary>
public string BoxLineCommentDelimiter {
get { return mFormatConfig.mBoxLineCommentDelimiter; }
}
/// <summary>
/// Prefix for non-unique address labels.
/// </summary>
public string NonUniqueLabelPrefix {
get { return mFormatConfig.mNonUniqueLabelPrefix; }
}
/// <summary>
/// When formatting a symbol with an offset, if this flag is set, generate code that
/// assumes the assembler applies the adjustment, then shifts the result. If not,
/// assume the assembler shifts the operand before applying the adjustment.
/// </summary>
public FormatConfig.ExpressionMode ExpressionMode {
get { return mFormatConfig.mExpressionMode; }
}
/// <summary>
/// Constructor. Initializes various fields based on the configuration. We want to
/// do as much work as possible here.
/// </summary>
public Formatter(FormatConfig config) {
mFormatConfig = config; // copy struct
mFormatConfig.Normalize();
if (string.IsNullOrEmpty(mFormatConfig.mNonUniqueLabelPrefix)) {
mFormatConfig.mNonUniqueLabelPrefix = "@";
}
if (mFormatConfig.mAddSpaceLongComment) {
mFullLineCommentDelimiterPlus = mFormatConfig.mFullLineCommentDelimiterBase + " ";
} else {
mFullLineCommentDelimiterPlus = mFormatConfig.mFullLineCommentDelimiterBase;
}
// Prep the static parts of the hex dump buffer.
mHexDumpBuffer = new char[73];
for (int i = 0; i < mHexDumpBuffer.Length; i++) {
mHexDumpBuffer[i] = ' ';
}
mHexDumpBuffer[6] = ':';
// Resolve boolean flags to character or string values.
if (mFormatConfig.mUpperHexDigits) {
mHexFmtChar = 'X';
} else {
mHexFmtChar = 'x';
}
if (mFormatConfig.mSuppressHexNotation) {
mHexPrefix = "";
} else {
mHexPrefix = "$";
}
if (mFormatConfig.mSuppressImpliedAcc) {
mAccChar = "";
} else if (mFormatConfig.mUpperOperandA) {
mAccChar = "A";
} else {
mAccChar = "a";
}
if (mFormatConfig.mUpperOperandXY) {
mXregChar = 'X';
mYregChar = 'Y';
} else {
mXregChar = 'x';
mYregChar = 'y';
}
if (mFormatConfig.mUpperOperandS) {
mSregChar = 'S';
} else {
mSregChar = 's';
}
for (int index = 0; index < 4; index++) {
int width = (index + 1) * 2;
mHexValueFormats[index] = mHexFmtChar + width.ToString();
}
// process the delimiter patterns
DelimiterSet chrDelim = mFormatConfig.mCharDelimiters;
if (chrDelim == null) {
Debug.WriteLine("NOTE: char delimiters not set");
chrDelim = DelimiterSet.GetDefaultCharDelimiters();
}
switch (mFormatConfig.mHexDumpCharConvMode) {
case FormatConfig.CharConvMode.Ascii:
mHexDumpCharConv = CharEncoding.ConvertAscii;
break;
case FormatConfig.CharConvMode.LowHighAscii:
mHexDumpCharConv = CharEncoding.ConvertLowAndHighAscii;
break;
case FormatConfig.CharConvMode.C64Petscii:
mHexDumpCharConv = CharEncoding.ConvertC64Petscii;
break;
case FormatConfig.CharConvMode.C64ScreenCode:
mHexDumpCharConv = CharEncoding.ConvertC64ScreenCode;
break;
default:
// most some things don't configure the hex dump; this is fine
mHexDumpCharConv = CharEncoding.ConvertLowAndHighAscii;
break;
}
}
/// <summary>
/// Formats a 24-bit offset value as hex.
/// </summary>
/// <param name="offset">Offset to format.</param>
/// <returns>Formatted string.</returns>
public string FormatOffset24(int offset) {
if (string.IsNullOrEmpty(mOffset24Format)) {
mOffset24Format = "+{0:" + mHexFmtChar + "6}";
}
return string.Format(mOffset24Format, offset & 0x0fffff);
}
/// <summary>
/// Formats a value in hexadecimal. The width is padded with zeroes to make the
/// length even (so it'll be $00, $0100, $010000, etc.) If minDigits is nonzero,
/// additional zeroes may be added.
/// </summary>
/// <param name="value">Value to format, up to 32 bits.</param>
/// <param name="minDigits">Minimum width, in printed digits (e.g. 4 is "0000").</param>
/// <returns>Formatted string.</returns>
public string FormatHexValue(int value, int minDigits) {
int width = minDigits > 2 ? minDigits : 2;
if (width < 8 && value > 0xffffff) {
width = 8;
} else if (width < 6 && value > 0xffff) {
width = 6;
} else if (width < 4 && value > 0xff) {
width = 4;
}
int index = (width / 2) - 1;
return mHexPrefix + value.ToString(mHexValueFormats[index]);
}
/// <summary>
/// Format a value as a number in the specified base.
/// </summary>
/// <param name="value">Value to format.</param>
/// <param name="numBase">Numeric base (2, 10, or 16).</param>
/// <returns>Formatted string.</returns>
public string FormatValueInBase(int value, int numBase) {
switch (numBase) {
case 2:
return FormatBinaryValue(value, 8);
case 10:
return FormatDecimalValue(value);
case 16:
return FormatHexValue(value, 2);
default:
Debug.Assert(false);
return "???";
}
}
/// <summary>
/// Formats a 32-bit integer value as unsigned decimal.
/// </summary>
/// <param name="value">Value to convert.</param>
/// <returns>Formatted string.</returns>
public string FormatDecimalValue(int value) {
return ((uint)value).ToString();
}
/// <summary>
/// Formats a value in binary, padding with zeroes so the length is a multiple of 8.
/// </summary>
/// <param name="value">Value to convert.</param>
/// <param name="minDigits">Minimum width, in printed digits. Will be rounded up to
/// a multiple of 8.</param>
/// <returns>Formatted string.</returns>
public string FormatBinaryValue(int value, int minDigits) {
string binaryStr = Convert.ToString(value, 2);
int desiredWidth = ((binaryStr.Length + 7) / 8) * 8;
if (desiredWidth < minDigits) {
desiredWidth = ((minDigits + 7) / 8) * 8;
}
return '%' + binaryStr.PadLeft(desiredWidth, '0');
}
/// <summary>
/// Formats a single-character operand. Output will be a delimited printable character
/// when possible, a hex value when the converted character is unprintable.
/// </summary>
/// <param name="value">Value to format. Could be a 16-bit immediate value.</param>
/// <param name="enc">Character encoding to use for value.</param>
/// <returns>Formatted string.</returns>
public string FormatCharacterValue(int value, CharEncoding.Encoding enc) {
if (value < 0 || value > 0xff) {
return FormatHexValue(value, 2);
}
DelimiterDef delimDef = mFormatConfig.mCharDelimiters.Get(enc);
if (delimDef == null) {
return FormatHexValue(value, 2);
}
string fmt = delimDef.FormatStr;
Debug.Assert(fmt != null);
CharEncoding.Convert conv;
switch (enc) {
case CharEncoding.Encoding.Ascii:
conv = CharEncoding.ConvertAscii;
break;
case CharEncoding.Encoding.HighAscii:
conv = CharEncoding.ConvertHighAscii;
break;
case CharEncoding.Encoding.C64Petscii:
conv = CharEncoding.ConvertC64Petscii;
break;
case CharEncoding.Encoding.C64ScreenCode:
conv = CharEncoding.ConvertC64ScreenCode;
break;
default:
return FormatHexValue(value, 2);
}
char ch = conv((byte)value);
if (ch == CharEncoding.UNPRINTABLE_CHAR || ch == delimDef.OpenDelim ||
ch == delimDef.CloseDelim) {
// We might be able to do better with delimiter clashes, e.g. '\'', but
// that's assembler-specific.
return FormatHexValue(value, 2);
} else {
// Possible optimization: replace fmt with a prefix/suffix pair, and just concat
return string.Format(fmt, ch);
}
}
/// <summary>
/// Formats a 16- or 24-bit address value. This is intended for the left column
/// of something (hex dump, code listing), not as an operand.
/// </summary>
/// <param name="address">Address to format.</param>
/// <param name="showBank">Set to true for CPUs with 24-bit address spaces.</param>
/// <returns>Formatted string.</returns>
public string FormatAddress(int address, bool showBank) {
if (mAddrFormatNoBank == null) {
mAddrFormatNoBank = "{0:" + mHexFmtChar + "4}";
mAddrFormatWithBank = "{0:" + mHexFmtChar + "2}/{1:" + mHexFmtChar + "4}";
}
if (showBank) {
return string.Format(mAddrFormatWithBank, address >> 16, address & 0xffff);
} else {
return string.Format(mAddrFormatNoBank, address & 0xffff);
}
}
/// <summary>
/// Formats a local variable label, prepending an identifying prefix if one has been
/// specified.
/// </summary>
public string FormatVariableLabel(string label) {
if (!string.IsNullOrEmpty(mFormatConfig.mLocalVariableLabelPrefix)) {
return mFormatConfig.mLocalVariableLabelPrefix + label;
} else {
return label;
}
}
/// <summary>
/// Formats an adjustment. Small values are formatted as "+decimal" or "-decimal",
/// larger values are formatted as hex. If no adjustment is required, an empty string
/// is returned.
/// </summary>
/// <param name="adjValue">Adjustment value.</param>
/// <returns>Formatted string.</returns>
public string FormatAdjustment(int adjValue) {
if (adjValue == 0) {
return string.Empty;
} else if (Math.Abs(adjValue) >= 256) {
// not using mHexPrefix here, since dec vs. hex matters
if (adjValue < 0) {
return "-$" + (-adjValue).ToString(mHexValueFormats[0]);
} else {
return "+$" + adjValue.ToString(mHexValueFormats[0]);
}
} else {
// This formats in decimal with a leading '+' or '-'. To avoid adding a plus
// on zero, we'd use "+#;-#;0", but we took care of the zero case above.
return adjValue.ToString("+0;-#");
}
}
/// <summary>
/// Formats the instruction opcode mnemonic, and caches the result.
///
/// It may be necessary to modify the mnemonic for some assemblers, e.g. LDA from a
/// 24-bit address might need to be LDAL, even if the high byte is nonzero.
/// </summary>
/// <param name="op">Opcode to format</param>
/// <param name="wdis">Width disambiguation specifier.</param>
/// <returns>Formatted string.</returns>
public string FormatOpcode(OpDef op, OpDef.WidthDisambiguation wdis) {
// TODO(someday): using op.Opcode as the key is a bad idea, as the operation may
// not be the same on different CPUs. We currently rely on the caller to discard
// the Formatter when the CPU definition changes. We'd be better off keying off of
// the OpDef object and factoring wdis in some other way.
int key = op.Opcode | ((int)wdis << 8);
if (!mOpcodeStrings.TryGetValue(key, out string opcodeStr)) {
// Not found, generate value.
opcodeStr = FormatMnemonic(op.Mnemonic, wdis);
// Memoize.
mOpcodeStrings[key] = opcodeStr;
}
return opcodeStr;
}
/// <summary>
/// Formats the string as an opcode mnemonic.
///
/// It may be necessary to modify the mnemonic for some assemblers, e.g. LDA from a
/// 24-bit address might need to be LDAL, even if the high byte is nonzero.
/// </summary>
/// <param name="mnemonic">Instruction mnemonic string.</param>
/// <param name="wdis">Width disambiguation specifier.</param>
/// <returns></returns>
public string FormatMnemonic(string mnemonic, OpDef.WidthDisambiguation wdis) {
string opcodeStr = mnemonic;
if (wdis == OpDef.WidthDisambiguation.ForceDirect) {
if (!string.IsNullOrEmpty(mFormatConfig.mForceDirectOpcodeSuffix)) {
opcodeStr += mFormatConfig.mForceDirectOpcodeSuffix;
}
} else if (wdis == OpDef.WidthDisambiguation.ForceAbs) {
if (!string.IsNullOrEmpty(mFormatConfig.mForceAbsOpcodeSuffix)) {
opcodeStr += mFormatConfig.mForceAbsOpcodeSuffix;
}
} else if (wdis == OpDef.WidthDisambiguation.ForceLong ||
wdis == OpDef.WidthDisambiguation.ForceLongMaybe) {
if (!string.IsNullOrEmpty(mFormatConfig.mForceLongOpcodeSuffix)) {
opcodeStr += mFormatConfig.mForceLongOpcodeSuffix;
}
} else {
Debug.Assert(wdis == OpDef.WidthDisambiguation.None);
}
if (mFormatConfig.mUpperOpcodes) {
opcodeStr = opcodeStr.ToUpperInvariant();
}
return opcodeStr;
}
/// <summary>
/// Generates an operand format.
/// </summary>
/// <param name="addrMode">Addressing mode.</param>
/// <param name="wdis">Width disambiguation mode.</param>
/// <returns>Format string.</returns>
private string GenerateOperandFormat(OpDef.AddressMode addrMode,
OpDef.WidthDisambiguation wdis) {
string fmt;
string wdisStr = string.Empty;
if (wdis == OpDef.WidthDisambiguation.ForceDirect) {
if (!string.IsNullOrEmpty(mFormatConfig.mForceDirectOperandPrefix)) {
wdisStr = mFormatConfig.mForceDirectOperandPrefix;
}
} else if (wdis == OpDef.WidthDisambiguation.ForceAbs) {
if (!string.IsNullOrEmpty(mFormatConfig.mForceAbsOperandPrefix)) {
wdisStr = mFormatConfig.mForceAbsOperandPrefix;
}
} else if (wdis == OpDef.WidthDisambiguation.ForceLong) {
if (!string.IsNullOrEmpty(mFormatConfig.mForceLongOperandPrefix)) {
wdisStr = mFormatConfig.mForceLongOperandPrefix;
}
} else if (wdis == OpDef.WidthDisambiguation.ForceLongMaybe) {
// Don't add a width disambiguator to an operand that is unambiguously long.
} else {
Debug.Assert(wdis == OpDef.WidthDisambiguation.None);
}
switch (addrMode) {
case AddressMode.Abs:
case AddressMode.AbsLong:
case AddressMode.BlockMove:
case AddressMode.StackAbs:
case AddressMode.DP:
case AddressMode.PCRel:
case AddressMode.PCRelLong: // BRL
case AddressMode.StackInt: // COP and two-byte BRK
case AddressMode.StackPCRelLong: // PER
case AddressMode.WDM:
fmt = wdisStr + "{0}";
break;
case AddressMode.AbsIndexX:
case AddressMode.AbsIndexXLong:
case AddressMode.DPIndexX:
fmt = wdisStr + "{0}," + mXregChar;
break;
case AddressMode.DPIndexY:
case AddressMode.AbsIndexY:
fmt = wdisStr + "{0}," + mYregChar;
break;
case AddressMode.AbsIndexXInd:
case AddressMode.DPIndexXInd:
fmt = wdisStr + "({0}," + mXregChar + ")";
break;
case AddressMode.AbsInd:
case AddressMode.DPInd:
case AddressMode.StackDPInd: // PEI
fmt = "({0})";
break;
case AddressMode.AbsIndLong:
case AddressMode.DPIndLong:
// IIgs monitor uses "()" for AbsIndLong, E&L says "[]". Assemblers
// seem to expect the latter.
fmt = "[{0}]";
break;
case AddressMode.Acc:
fmt = mAccChar;
break;
case AddressMode.DPIndIndexY:
fmt = "({0})," + mYregChar;
break;
case AddressMode.DPIndIndexYLong:
fmt = "[{0}]," + mYregChar;
break;
case AddressMode.Imm:
case AddressMode.ImmLongA:
case AddressMode.ImmLongXY:
fmt = "#{0}";
break;
case AddressMode.Implied:
case AddressMode.StackPull:
case AddressMode.StackPush:
case AddressMode.StackRTI:
case AddressMode.StackRTL:
case AddressMode.StackRTS:
fmt = string.Empty;
break;
case AddressMode.StackRel:
fmt = "{0}," + mSregChar;
break;
case AddressMode.StackRelIndIndexY:
fmt = "({0}," + mSregChar + ")," + mYregChar;
break;
case AddressMode.Unknown:
default:
Debug.Assert(false);
fmt = "???";
break;
}
return fmt;
}
/// <summary>
/// Formats the instruction operand.
/// </summary>
/// <param name="op">Opcode definition (needed for address mode).</param>
/// <param name="contents">Label or numeric operand value.</param>
/// <param name="wdis">Width disambiguation value.</param>
/// <returns>Formatted string.</returns>
public string FormatOperand(OpDef op, string contents, OpDef.WidthDisambiguation wdis) {
Debug.Assert(((int)op.AddrMode & 0xff) == (int) op.AddrMode);
int key = (int) op.AddrMode | ((int)wdis << 8);
if (!mOperandFormats.TryGetValue(key, out string format)) {
format = mOperandFormats[key] = GenerateOperandFormat(op.AddrMode, wdis);
}
return string.Format(format, contents);
}
/// <summary>
/// Formats a pseudo-opcode.
/// </summary>
/// <param name="opstr">Pseudo-op string to format.</param>
/// <returns>Formatted string.</returns>
public string FormatPseudoOp(string opstr) {
if (!mPseudoOpStrings.TryGetValue(opstr, out string result)) {
if (mFormatConfig.mUpperPseudoOpcodes) {
result = mPseudoOpStrings[opstr] = opstr.ToUpperInvariant();
} else {
result = mPseudoOpStrings[opstr] = opstr;
}
}
return result;
}
/// <summary>
/// Generates a format string for N hex bytes.
/// </summary>
/// <param name="len">Number of bytes to handle in the format.</param>
private void GenerateByteFormat(int len) {
Debug.Assert(len <= MAX_BYTE_DUMP);
StringBuilder sb = new StringBuilder(len * 7);
for (int i = 0; i < len; i++) {
if (i != 0 && mFormatConfig.mSpacesBetweenBytes) {
sb.Append(' ');
}
// e.g. "{0:x2}"
sb.Append("{" + i + ":" + mHexFmtChar + "2}");
}
mByteDumpFormats[len - 1] = sb.ToString();
}
/// <summary>
/// Formats 1-4 bytes as hex values.
/// </summary>
/// <param name="data">Data source.</param>
/// <param name="offset">Start offset within data array.</param>
/// <param name="length">Number of bytes to print. Fewer than this many may
/// actually appear.</param>
/// <returns>Formatted data string.</returns>
public string FormatBytes(byte[] data, int offset, int length) {
Debug.Assert(length > 0);
int printLen = length < MAX_BYTE_DUMP ? length : MAX_BYTE_DUMP;
if (string.IsNullOrEmpty(mByteDumpFormats[printLen - 1])) {
GenerateByteFormat(printLen);
}
string format = mByteDumpFormats[printLen - 1];
string result;
// The alternative is to allocate a temporary object[] and copy the integers
// into it, which requires boxing. We know we're only printing 1-4 bytes, so
// it's easier to just handle each case individually.
switch (printLen) {
case 1:
result = string.Format(format, data[offset]);
break;
case 2:
result = string.Format(format, data[offset], data[offset + 1]);
break;
case 3:
result = string.Format(format,
data[offset], data[offset + 1], data[offset + 2]);
break;
case 4:
result = string.Format(format,
data[offset], data[offset + 1], data[offset + 2], data[offset + 3]);
break;
default:
result = "INTERNAL ERROR";
break;
}
if (length > printLen) {
result += "...";
}
return result;
}
/// <summary>
/// Formats an end-of-line comment, prepending an end-of-line comment delimiter.
/// </summary>
/// <param name="comment">Comment string; may be empty.</param>
/// <returns>Formatted string.</returns>
public string FormatEolComment(string comment) {
if (string.IsNullOrEmpty(comment) ||
string.IsNullOrEmpty(mFormatConfig.mEndOfLineCommentDelimiter)) {
return comment;
} else {
return mFormatConfig.mEndOfLineCommentDelimiter + comment;
}
}
/// <summary>
/// Formats a collection of bytes as a dense hex string.
/// </summary>
/// <param name="data">Data source.</param>
/// <param name="offset">Start offset within data array.</param>
/// <param name="length">Number of bytes to print.</param>
/// <returns>Formatted data string.</returns>
public string FormatDenseHex(byte[] data, int offset, int length) {
char[] hexChars = mFormatConfig.mUpperHexDigits ? sHexCharsUpper : sHexCharsLower;
char[] text;
if (mFormatConfig.mCommaSeparatedDense) {
text = new char[length * 4 - 1];
for (int i = 0; i < length; i++) {
byte val = data[offset + i];
text[i * 4] = '$';
text[i * 4 + 1] = hexChars[val >> 4];
text[i * 4 + 2] = hexChars[val & 0x0f];
if (i != length - 1) {
text[i * 4 + 3] = ',';
}
}
} else {
text = new char[length * 2];
for (int i = 0; i < length; i++) {
byte val = data[offset + i];
text[i * 2] = hexChars[val >> 4];
text[i * 2 + 1] = hexChars[val & 0x0f];
}
}
return new string(text);
}
/// <summary>
/// Returns the number of characters output for each byte when formatting dense hex.
/// </summary>
/// <remarks>
/// This isn't quite right, because you don't need a comma after the very last element
/// in the list for comma-separated values. Handling this correctly for multi-line
/// items is more trouble than it's worth though.
/// </remarks>
public int CharsPerDenseByte {
get {
if (mFormatConfig.mCommaSeparatedDense) {
return 4;
} else {
return 2;
}
}
}
/// <summary>
/// Formats up to 16 bytes of data into a single line hex dump, in this format:
/// <pre>012345: 00 11 22 33 44 55 66 77 88 99 aa bb cc dd ee ff 0123456789abcdef</pre>
/// </summary>
/// <param name="data">Reference to data.</param>
/// <param name="offset">Start offset.</param>
/// <returns>Formatted string.</returns>
public string FormatHexDump(byte[] data, int offset) {
int length = Math.Min(16, data.Length - offset);
FormatHexDumpCommon(data, offset, offset, length);
// this is the only allocation
return new string(mHexDumpBuffer);
}
/// <summary>
/// Formats up to 16 bytes of data into a single line hex dump. The output is
/// appended to the StringBuilder.
/// </summary>
/// <param name="data">Reference to data.</param>
/// <param name="offset">Start offset.</param>
/// <param name="sb">StringBuilder that receives output.</param>
public void FormatHexDump(byte[] data, int offset, StringBuilder sb) {
int length = Math.Min(16, data.Length - offset);
FormatHexDumpCommon(data, offset, offset, length);
sb.Append(mHexDumpBuffer);
}
public void FormatHexDump(byte[] data, int offset, int addr, int length,
StringBuilder sb) {
FormatHexDumpCommon(data, offset, addr, length);
sb.Append(mHexDumpBuffer);
}
/// <summary>
/// Formats up to 16 bytes of data into mHexDumpBuffer.
/// </summary>
private void FormatHexDumpCommon(byte[] data, int offset, int addr, int length) {
Debug.Assert(offset >= 0 && offset < data.Length);
Debug.Assert(data.Length < (1 << 24));
const int dataCol = 8;
const int asciiCol = 57;
char[] hexChars = mFormatConfig.mUpperHexDigits ? sHexCharsUpper : sHexCharsLower;
char[] outBuf = mHexDumpBuffer;
int skip = addr & 0x0f; // we skip this many entries...
offset -= skip; // ...so adjust offset to balance it
addr &= ~0x0f;
// address field
for (int i = 5; i >= 0; i--) {
outBuf[i] = hexChars[addr & 0x0f];
addr >>= 4;
}
// If addr doesn't start at xxx0, pad it.
int index;
for (index = 0; index < skip; index++) {
outBuf[dataCol + index * 3] = outBuf[dataCol + index * 3 + 1] =
outBuf[asciiCol + index] = ' ';
}
// hex digits and characters
for (int i = 0; i < length; i++) {
byte val = data[offset + index];
outBuf[dataCol + index * 3] = hexChars[val >> 4];
outBuf[dataCol + index * 3 + 1] = hexChars[val & 0x0f];
outBuf[asciiCol + index] = CharConv(val);
index++;
}
// for partial line, clear out previous contents
for (; index < 16; index++) {
outBuf[dataCol + index * 3] =
outBuf[dataCol + index * 3 + 1] =
outBuf[asciiCol + index] = ' ';
}
}
/// <summary>
/// Converts a byte into printable form according to the current hex dump
/// character conversion mode.
/// </summary>
/// <param name="val">Value to convert.</param>
/// <returns>Printable character.</returns>
private char CharConv(byte val) {
char ch = mHexDumpCharConv(val);
if (ch != CharEncoding.UNPRINTABLE_CHAR) {
return ch;
} else if (mFormatConfig.mHexDumpAsciiOnly) {
return '.';
} else {
// Certain values make the hex dump ListView freak out in WinForms, but work
// fine in WPF. The "control pictures" are a nice idea, but in practice they're
// unreadably small and provide no benefit. The black-diamond "replacement
// character" is dark and makes everything feel noisy. Middle-dot is subtle,
// but sufficiently different from a '.' to be useful.
//if (ch < 0x20) {
// return (char)(ch + '\u2400'); // Unicode "control pictures" block
//}
//return '\ufffd'; // Unicode "replacement character"
//return '\u00bf'; // INVERTED QUESTION MARK
return '\u00b7'; // MIDDLE DOT
}
}
}
}