1
0
mirror of https://github.com/fadden/6502bench.git synced 2024-11-04 15:05:03 +00:00
6502bench/SourceGen/LineListGen.cs
Andy McFadden 0fb0b4cca8 Fix some address region issues
Fixed a crash when the listing is refreshed while a .adrend line is
the only thing selected.

Fixed .adrend lines being spammed if the last thing before it is a
multi-byte item, and you edit a comment / note / label on that line.
Harmless, but weird.

Also, added keyboard shortcuts in DefSymbol editor, so you can change
from address to constant with Alt+C.
2021-10-17 18:02:29 -07:00

1748 lines
82 KiB
C#

/*
* Copyright 2019 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.Windows.Media;
using System.Text;
using Asm65;
using CommonUtil;
using FormattedParts = SourceGen.DisplayList.FormattedParts;
using TextScanMode = SourceGen.ProjectProperties.AnalysisParameters.TextScanMode;
namespace SourceGen {
/// <summary>
/// Converts file data and Anattrib contents into a series of strings and format metadata.
/// </summary>
public class LineListGen {
/// <summary>
/// Color multiplier for Notes. Used for "dark" mode.
/// </summary>
public float NoteColorMultiplier { get; set; } = 1.0f;
/// <summary>
/// List of display lines.
/// </summary>
private List<Line> mLineList;
/// <summary>
/// List of formatted parts to be presented to the user. This has one entry per line.
/// </summary>
/// <remarks>
/// Separating FormattedParts out of Line seems odd at first, but we need changes to
/// DisplayList to cause events in XAML. I'm thinking the artificial separation of
/// Line from the formatted data holder may make future ports easier.
/// </remarks>
private DisplayList mDisplayList;
/// <summary>
/// Project that contains the data we're formatting, notably the FileData and
/// Anattribs arrays.
/// </summary>
private DisasmProject mProject;
/// <summary>
/// Code/data formatter.
/// </summary>
private Formatter mFormatter;
/// <summary>
/// If set, prepend cycle counts to EOL comments.
/// </summary>
private bool mShowCycleCounts;
/// <summary>
/// Names for pseudo-ops.
/// </summary>
private PseudoOp.PseudoOpNames mPseudoOpNames;
/// <summary>
/// Cache of previously-formatted data. The data is stored with references to
/// dependencies, so it should not be necessary to explicitly clear this.
/// </summary>
private FormattedOperandCache mFormattedLineCache;
/// <summary>
/// Local variable table data extractor.
/// </summary>
private LocalVariableLookup mLvLookup;
/// <summary>
/// Character test, for display of character data next to default data items.
/// </summary>
private CharEncoding.InclusionTest mCharTest;
private CharEncoding.Convert mCharConv;
/// <summary>
/// One of these per line of output in the display. It should be possible to draw
/// all of the output without needing to refer back to the project data. (Currently
/// making an exception for some selection-dependent field highlighting.)
///
/// Base fields are immutable, but the Parts property is set after creation.
/// </summary>
public class Line {
// Extremely-negative offset value ensures it's at the very top.
public const int HEADER_COMMENT_OFFSET = int.MinValue + 1;
// These need to be bit flags so we can record which parts associated with a
// given offset are selected.
[FlagsAttribute]
public enum Type {
Unclassified = 0,
// Primary functional items.
Code = 1 << 0,
Data = 1 << 1, // includes inline data
CodeOrData = (Code | Data),
// Decorative items, added by user or formatter.
LongComment = 1 << 2,
Note = 1 << 3,
Blank = 1 << 4,
// Assembler directives.
ArStartDirective = 1 << 5,
ArEndDirective = 1 << 6,
EquDirective = 1 << 7,
RegWidthDirective = 1 << 8,
DataBankDirective = 1 << 9,
// Additional metadata.
LocalVariableTable = 1 << 10,
VisualizationSet = 1 << 11,
}
/// <summary>
/// Line type.
/// </summary>
public Type LineType { get; private set; }
/// <summary>
/// Numeric offset value. Used to map a line item to the Anattrib. Note this is
/// set for all lines, and is the same for all lines in a multi-line sequence,
/// e.g. every line in a long comment has the file offset with which it is associated.
/// </summary>
public int FileOffset { get; private set; }
/// <summary>
/// Number of offsets this line covers. Will be > 0 for code and data, zero for
/// everything else. The same value is used for all lines in a multi-line sequence.
/// </summary>
public int OffsetSpan { get; private set; }
/// <summary>
/// For multi-line entries, this indicates which line is represented. For
/// single-line entries, this will be zero.
/// </summary>
public int SubLineIndex { get; private set; }
/// <summary>
/// Strings for display. Creation may be deferred. Use the DisplayList
/// GetFormattedParts() method to access this property.
/// </summary>
/// <remarks>
/// Certain elements, such as multi-line comments, must be formatted to determine
/// the number of lines they span. We retain the results to avoid formatting
/// them twice.
/// </remarks>
public FormattedParts Parts { get; set; }
/// <summary>
/// Background color, used for Notes.
/// </summary>
public Color BackgroundColor { get; set; }
/// <summary>
/// String for searching. May be created on demand when the Line is first searched.
/// </summary>
public string SearchString { get; set; }
public Line(int offset, int span, Type type) : this(offset, span, type, 0) { }
public Line(int offset, int span, Type type, int subLineIndex) {
FileOffset = offset;
OffsetSpan = span;
LineType = type;
SubLineIndex = subLineIndex;
}
/// <summary>
/// True if this line is code or data.
/// </summary>
public bool IsCodeOrData {
get {
return LineType == Type.Code || LineType == Type.Data;
}
}
/// <summary>
/// True if this line is an address range start or end.
/// </summary>
public bool IsAddressRangeDirective {
get {
return LineType == Type.ArStartDirective || LineType == Type.ArEndDirective;
}
}
/// <summary>
/// Returns true if the specified offset is represented by this line. There
/// will be only one code/data line for a given offset, but there may be
/// multiple others (comments, notes, etc.) associated with it.
/// </summary>
/// <param name="offset"></param>
/// <returns></returns>
public bool Contains(int offset) {
// Note OffsetSpan can be zero.
return (offset == FileOffset ||
(offset >= FileOffset && offset < FileOffset + OffsetSpan));
}
public override string ToString() {
return "Line type=" + LineType + " off=+" + FileOffset.ToString("x6") +
" span=" + OffsetSpan;
}
}
/// <summary>
/// Captures the set of selected lines. Lines are identified by offset and type.
///
/// The idea is to save the selection, rebuild the list -- potentially moving
/// stuff around -- and then rebuild the selection bitmap by finding matching
/// items.
///
/// We don't try to identify parts of multi-line things. If you've selected
/// part of a multi-line string, then when we restore the selection you'll have
/// the entire string selected. For the operations that are possible across
/// multiple offsets, this seems like reasonable behavior.
///
/// We can't precisely restore the selection in terms of which file offsets
/// are selected. If you select one byte and apply an analyzer tag, we'll restore
/// the selection to a line with 1-4 bytes. This gets weird if you hit "undo",
/// as you will then have 1-4 bytes selected rather than the original one. It
/// might be better to just clear the selection on "undo".
/// </summary>
public class SavedSelection {
private class Tag {
public int mOffset;
public int mSpan;
public Line.Type mTypes;
public Tag(int offset, int span, Line.Type lineType) {
//Debug.Assert(offset >= 0);
Debug.Assert(span >= 0);
mOffset = offset;
mSpan = (span == 0) ? 1 : span;
mTypes = lineType;
}
}
private List<Tag> mSelectionTags = new List<Tag>();
/// <summary>
/// This is a place to save the file offset associated with the ListView's
/// TopItem, so we can position the list appropriately.
/// </summary>
private NavStack.Location mTopPosition;
// Use Generate().
private SavedSelection() { }
/// <summary>
/// Creates a new SavedSelection object, generating a list of tags from the
/// lines that are currently selected.
///
/// If nothing is selected, SavedSelection will have no members.
/// </summary>
/// <param name="dl">Display list, with list of Lines.</param>
/// <param name="sel">Bit vector specifying which lines are selected.</param>
/// <param name="topIndex">Index of line that appears at the top of the list
/// control.</param>
/// <returns>New SavedSelection object.</returns>
public static SavedSelection Generate(LineListGen dl, DisplayListSelection sel,
int topIndex) {
SavedSelection savedSel = new SavedSelection();
//Debug.Assert(topOffset >= 0);
int topOffset = dl[topIndex].FileOffset;
int firstIndex = dl.FindLineIndexByOffset(topOffset);
Debug.Assert(topIndex >= firstIndex);
savedSel.mTopPosition = new NavStack.Location(topOffset, topIndex - firstIndex,
NavStack.GoToMode.JumpToAdjIndex);
List<Line> lineList = dl.mLineList;
Debug.Assert(lineList.Count == sel.Length);
// Generate tags, which are a combination of the offset, span, and a merge
// of types of all the lines associated with that offset.
//
// We may want to consider some sort of optimization for a "select all"
// operation, although there aren't many changes you can make after selecting
// all lines in a very large file.
Tag tag = null;
int curOffset = -1;
for (int i = 0; i < lineList.Count; i++) {
if (!sel[i]) {
continue;
}
Line line = lineList[i];
// Code start/stop tags can transform code to data and vice-versa, so we
// want the tag to reflect the fact that both could exist.
Line.Type lineType = line.LineType;
if (lineType == Line.Type.Code || lineType == Line.Type.Data) {
lineType = Line.Type.CodeOrData;
}
if (line.FileOffset == curOffset ||
(tag != null && lineType == Line.Type.ArEndDirective)) {
// Another item at same offset. We special-case the arend directive
// because it's contained within the previous item, so we want it to be
// set on the existing [offset,offset+span) range tag.
tag.mSpan = Math.Max(tag.mSpan, line.OffsetSpan);
tag.mTypes |= lineType;
Debug.Assert(line.FileOffset >= tag.mOffset &&
line.FileOffset < tag.mOffset + tag.mSpan);
} else {
// advanced to new offset, flush previous
if (tag != null) {
savedSel.mSelectionTags.Add(tag);
}
curOffset = line.FileOffset;
tag = new Tag(line.FileOffset, line.OffsetSpan, lineType);
}
}
if (curOffset == -1) {
// It's hard to cause an action that requires save/restore when you don't
// have anything selected in the ListView. However, this can happen if
// you do a sequence like:
// - Open a file that starts with a JMP followed by data.
// - Click on the blank line below the code, which has the code's offset,
// and select "remove atags". This causes the blank line to vanish,
// so the Restore() won't select anything.
// - Click "undo".
Debug.WriteLine("NOTE: no selection found");
} else {
// Add the in-progress tag to the list.
savedSel.mSelectionTags.Add(tag);
}
return savedSel;
}
/// <summary>
/// Creates a selection set by identifying the set of lines in the display list
/// that correspond to items in the SavedSelection tag list.
/// </summary>
/// <param name="dl">Display list, with list of Lines.</param>
/// <param name="topIndex">Index of line to place at top of list control.</param>
/// <returns>Set of selected lines.</returns>
public DisplayListSelection Restore(LineListGen dl, out int topIndex) {
List<Line> lineList = dl.mLineList;
DisplayListSelection sel = new DisplayListSelection(lineList.Count);
topIndex = -1;
// Walk through the tag list, which is ordered by ascending offset, and
// through the display list, which is similarly ordered.
int tagIndex = 0;
int lineIndex = 0;
while (tagIndex < mSelectionTags.Count && lineIndex < lineList.Count) {
Tag tag = mSelectionTags[tagIndex];
int lineOffset = lineList[lineIndex].FileOffset;
// If a line encompassing this offset was at the top of the ListView
// control before, use this line's index as the top.
if (topIndex < 0 && lineList[lineIndex].Contains(mTopPosition.Offset)) {
topIndex = lineIndex;
}
if (lineOffset >= tag.mOffset && lineOffset < tag.mOffset + tag.mSpan) {
// Intersection. If the line type matches, add it to the set.
// TODO(someday): this is doing the wrong thing when we have more
// than one blank line at an offset.
if ((tag.mTypes & lineList[lineIndex].LineType) != 0) {
sel[lineIndex] = true;
}
// Advance to the next line entry.
lineIndex++;
} else if (tag.mOffset < lineOffset) {
// advance tag
tagIndex++;
} else {
Debug.Assert(tag.mOffset > lineOffset);
lineIndex++;
}
}
// Continue search for topIndex, if necessary.
while (topIndex < 0 && lineIndex < lineList.Count) {
if (lineList[lineIndex].Contains(mTopPosition.Offset)) {
topIndex = lineIndex;
break;
}
lineIndex++;
}
Debug.WriteLine("TopOffset " + mTopPosition + " --> index " + topIndex);
// Adjust position within an element. This is necessary so we don't jump to
// the top of multi-line long comments or notes whenever any part of that
// comment or note is at the top of the list.
if (topIndex >= 0 && mTopPosition.LineDelta > 0) {
int adjIndex = topIndex + mTopPosition.LineDelta;
if (adjIndex >= lineList.Count ||
lineList[adjIndex].FileOffset != mTopPosition.Offset) {
Debug.WriteLine("Can't adjust top position");
// can't adjust; maybe they deleted several lines from comment
} else {
topIndex = adjIndex;
Debug.WriteLine("Top index adjusted to " + adjIndex);
}
}
if (topIndex < 0) {
// This can happen if you delete the header comment while scrolled
// to the top of the list.
topIndex = 0;
}
// If the selected item was deleted, we can't restore the selection. This is
// annoying when you're using the keyboard, because if you hit Del to delete a
// note and then hit down-arrow you jump to the top of the file. If we detect
// this situation, we want to select the thing that appeared below it. We can
// approximate this by just selecting the first thing at the offset (which is
// not quite right if you have both a long comment and a note, but it'll do).
//
// This doesn't help for some of the header stuff, e.g. if you select the last
// platform/project equate and then remove one symbol, the selection gets lost.
if (sel.Count == 0 && mSelectionTags.Count != 0) {
Debug.WriteLine("Lost selection");
int tryOffset = mSelectionTags[0].mOffset;
int tryIndex = LineListGen.FindLineByOffset(lineList, tryOffset);
if (tryIndex >= 0) {
Debug.WriteLine(" Setting to offset +" + tryOffset.ToString("x6") +
" (line " + tryIndex + ")");
sel[tryIndex] = true;
}
}
return sel;
}
public void DebugDump() {
Debug.WriteLine("Selection (" + mSelectionTags.Count + " offsets):");
foreach (Tag tag in mSelectionTags) {
Debug.WriteLine(" +" + tag.mOffset.ToString("x6") + "/" +
tag.mSpan + ": " + tag.mTypes);
}
}
}
/// <summary>
/// Constructor.
/// </summary>
/// <param name="proj">Project object.</param>
/// <param name="formatter">Formatter object.</param>
public LineListGen(DisasmProject proj, DisplayList displayList, Formatter formatter,
PseudoOp.PseudoOpNames opNames) {
Debug.Assert(proj != null);
Debug.Assert(displayList != null);
Debug.Assert(formatter != null);
Debug.Assert(opNames != null);
mProject = proj;
mDisplayList = displayList;
mFormatter = formatter;
mPseudoOpNames = opNames;
mLineList = new List<Line>();
mFormattedLineCache = new FormattedOperandCache();
mShowCycleCounts = AppSettings.Global.GetBool(AppSettings.FMT_SHOW_CYCLE_COUNTS,
false);
mLvLookup = new LocalVariableLookup(mProject.LvTables, mProject, null, false, false);
mDisplayList.ListGen = this;
}
/// <summary>
/// Changes the Formatter object. Clears the line list, instigating a full re-render.
/// </summary>
/// <param name="formatter">Formatter object.</param>
public void SetFormatter(Formatter formatter) {
mFormatter = formatter;
mLineList.Clear();
mDisplayList.Clear();
// We probably just changed settings, so update this as well.
mShowCycleCounts = AppSettings.Global.GetBool(AppSettings.FMT_SHOW_CYCLE_COUNTS,
false);
}
/// <summary>
/// Changes the pseudo-op name object. Clears the line list, instigating a
/// full re-render. If the new set is unchanged from the old set, nothing is done.
/// </summary>
/// <param name="opNames">Pseudo-op names.</param>
public void SetPseudoOpNames(PseudoOp.PseudoOpNames opNames) {
if (mPseudoOpNames == opNames) {
return;
}
mPseudoOpNames = opNames;
Debug.Assert(mPseudoOpNames == opNames);
mLineList.Clear();
mDisplayList.Clear();
}
/// <summary>
/// Number of lines in the list.
/// </summary>
public int Count { get { return mLineList.Count; } }
/// <summary>
/// Retrieves the Nth element.
/// </summary>
public Line this[int key] {
get {
return mLineList[key];
}
}
/// <summary>
/// Returns the Line's FormattedParts object, generating it first if necessary.
/// </summary>
/// <returns>Object with formatted strings.</returns>
public FormattedParts GetFormattedParts(int index) {
Line line = mLineList[index];
if (line.Parts == null) {
FormattedParts parts;
switch (line.LineType) {
case Line.Type.Code:
parts = GenerateInstructionLine(line.FileOffset, line.OffsetSpan);
break;
case Line.Type.Data:
parts = GenerateDataLine(line.FileOffset, line.SubLineIndex);
break;
case Line.Type.LocalVariableTable:
parts = GenerateLvTableLine(line.FileOffset, line.SubLineIndex);
break;
case Line.Type.VisualizationSet:
mProject.VisualizationSets.TryGetValue(line.FileOffset,
out VisualizationSet visSet);
parts = FormattedParts.CreateVisualizationSet(visSet);
break;
case Line.Type.Blank:
// Nothing to do.
parts = FormattedParts.CreateBlankLine();
break;
case Line.Type.ArStartDirective:
case Line.Type.ArEndDirective:
case Line.Type.RegWidthDirective:
case Line.Type.DataBankDirective:
case Line.Type.LongComment:
case Line.Type.Note:
// should have been done already
default:
Debug.Assert(false);
parts = FormattedParts.Create("x", "x", "x", "x", "x", "x", "x", "x", "x",
FormattedParts.PartFlags.None);
break;
}
line.Parts = parts;
}
return line.Parts;
}
/// <summary>
/// Returns a string with the concatenation of the searchable portions of the line.
/// Different sections are separated with an unlikely unicode character. The goal
/// is to have a single string per line that can be searched quickly, without having
/// adjacent fields spill into each other.
/// </summary>
/// <param name="index">Line index.</param>
/// <returns>Formatted line contents.</returns>
public string GetSearchString(int index) {
Line line = mLineList[index];
if (line.SearchString == null) {
const char sep = '\u203b'; // REFERENCE MARK
FormattedParts parts = GetFormattedParts(index);
StringBuilder sb = new StringBuilder();
// Some parts may be null, e.g. for long comments. Append() can deal.
sb.Append(parts.Label);
sb.Append(sep);
sb.Append(parts.Opcode);
sb.Append(sep);
sb.Append(parts.Operand);
sb.Append(sep);
sb.Append(parts.Comment);
line.SearchString = sb.ToString();
}
return line.SearchString;
}
/// <summary>
/// Finds the first line entry that encompasses the specified offset.
/// </summary>
/// <param name="offset">Offset to search for. Negative values are allowed.</param>
/// <returns>Line list index, or -1 if not found.</returns>
private static int FindLineByOffset(List<Line> lineList, int offset) {
if (lineList.Count == 0) {
return -1;
}
// Start with a binary search.
int low = 0;
int high = lineList.Count - 1;
int mid = -1;
bool found = false;
while (low <= high) {
mid = (low + high) / 2;
Line line = lineList[mid];
if (line.Contains(offset)) {
// found a match
found = true;
break;
} else if (line.FileOffset > offset) {
// too big, move the high end in
high = mid - 1;
} else if (line.FileOffset < offset) {
// too small, move the low end in
low = mid + 1;
} else {
// WTF
throw new Exception("Bad binary search");
}
}
if (!found) {
return -1;
}
// We found *a* matching line. Seek backward to find the *first* matching line.
while (mid > 0) {
Line upLine = lineList[mid - 1];
if (upLine.Contains(offset)) {
mid--;
} else {
break;
}
}
return mid;
}
/// <summary>
/// Finds the first line entry that encompasses the specified offset.
/// </summary>
/// <param name="offset">Offset to search for.</param>
/// <returns>Line list index, or -1 if not found.</returns>
public int FindLineIndexByOffset(int offset) {
return FindLineByOffset(mLineList, offset);
}
/// <summary>
/// Finds the code or data line entry that encompasses the specified offset.
/// </summary>
/// <param name="offset">Offset to search for.</param>
/// <returns>Line list index, or -1 if not found.</returns>
public int FindCodeDataIndexByOffset(int offset) {
if (offset < 0) {
// Header offset. No code or data here.
return -1;
}
int index = FindLineByOffset(mLineList, offset);
if (index < 0) {
return -1;
}
while (mLineList[index].LineType != Line.Type.Code &&
mLineList[index].LineType != Line.Type.Data) {
index++;
}
return index;
}
/// <summary>
/// Generates Lines for the entire project.
/// </summary>
public void GenerateAll() {
// Do this now in case the project properties have changed.
ConfigureCharacterEncoding();
mLineList.Clear();
List<Line> headerLines = GenerateHeaderLines(mProject, mFormatter, mPseudoOpNames);
mLineList.InsertRange(0, headerLines);
GenerateLineList(0, mProject.FileData.Length - 1, mLineList);
mDisplayList.ResetList(mLineList.Count);
Debug.Assert(ValidateLineList(), "Display list failed validation");
}
private void ConfigureCharacterEncoding() {
TextScanMode textMode = mProject.ProjectProps.AnalysisParams.DefaultTextScanMode;
switch (textMode) {
case TextScanMode.LowAscii:
mCharTest = CharEncoding.IsPrintableAscii;
mCharConv = CharEncoding.ConvertAscii;
break;
case TextScanMode.LowHighAscii:
mCharTest = CharEncoding.IsPrintableLowOrHighAscii;
mCharConv = CharEncoding.ConvertLowAndHighAscii;
break;
case TextScanMode.C64Petscii:
mCharTest = CharEncoding.IsPrintableC64Petscii;
mCharConv = CharEncoding.ConvertC64Petscii;
break;
case TextScanMode.C64ScreenCode:
mCharTest = CharEncoding.IsPrintableC64ScreenCode;
mCharConv = CharEncoding.ConvertC64ScreenCode;
break;
default:
Debug.Assert(false);
mCharTest = CharEncoding.IsPrintableAscii;
mCharConv = CharEncoding.ConvertAscii;
break;
}
}
/// <summary>
/// Generates a list of Lines for the specified range of offsets, replacing
/// existing values.
/// </summary>
/// <param name="startOffset">First offset. Must be the start of an instruction
/// or data area.</param>
/// <param name="endOffset">End offset (inclusive).</param>
public void GenerateRange(int startOffset, int endOffset) {
if (startOffset < 0) {
// Clear existing header lines. Find the first non-header item.
int headerEndIndex = FindLineByOffset(mLineList, 0);
if (headerEndIndex == 0) {
// no header lines present
Debug.WriteLine("No header lines found");
} else {
Debug.WriteLine("Removing " + headerEndIndex + " header lines");
mLineList.RemoveRange(0, headerEndIndex);
}
List<Line> headerLines = GenerateHeaderLines(mProject, mFormatter, mPseudoOpNames);
mLineList.InsertRange(0, headerLines);
mDisplayList.ClearListSegment(0, headerEndIndex, headerLines.Count);
if (endOffset < 0) {
// nothing else to do -- header-only change
return;
}
// do the rest
startOffset = 0;
}
Debug.Assert(startOffset >= 0);
Debug.Assert(endOffset < mProject.FileData.Length);
Debug.Assert(endOffset >= startOffset);
//Debug.WriteLine("DL gen range [" + startOffset + "," + endOffset + "]");
// Find the start index. The start offset should always appear at the
// start of a Line (as opposed to being in the middle of a multi-byte instruction)
// because it comes from item selection.
int startIndex = FindLineByOffset(mLineList, startOffset);
if (startIndex < 0) {
Debug.Assert(false, "Unable to find startOffset " + startOffset);
GenerateAll();
return;
}
// Find the end index. The end offset can be part of a multi-line data item, like
// a long string. Find the first Line that starts at an offset larger than endOffset.
int endIndex;
if (startOffset == endOffset) {
// Simple optimization for single-offset groups.
endIndex = startIndex;
} else {
endIndex = FindLineByOffset(mLineList, endOffset);
}
if (endIndex < 0) {
Debug.Assert(false, "Unable to find endOffset " + endOffset);
GenerateAll();
return;
}
// There may be more than one line involved, so we need to scan forward.
for (endIndex++; endIndex < mLineList.Count; endIndex++) {
if (mLineList[endIndex].FileOffset > endOffset) {
endIndex--;
break;
}
}
if (endIndex == mLineList.Count) {
// whoops, loop ended before we had a chance to decrement
endIndex = mLineList.Count - 1;
}
Debug.WriteLine("GenerateRange: offset [+" + startOffset.ToString("x6") + ",+" +
endOffset.ToString("x6") +
"] maps to index [" + startIndex + "," + endIndex + "]");
Debug.Assert(endIndex >= startIndex);
// Create temporary list to hold new lines. Set the initial capacity to
// the previous size, on the assumption that it won't change much.
List<Line> newLines = new List<Line>(endIndex - startIndex + 1);
GenerateLineList(startOffset, endOffset, newLines);
// Out with the old, in with the new.
mLineList.RemoveRange(startIndex, endIndex - startIndex + 1);
mLineList.InsertRange(startIndex, newLines);
mDisplayList.ClearListSegment(startIndex, endIndex - startIndex + 1, newLines.Count);
Debug.Assert(ValidateLineList(), "Display list failed validation");
}
/// <summary>
/// Validates the line list, confirming that every offset is represented exactly once.
/// </summary>
/// <returns>True if all is well.</returns>
private bool ValidateLineList() {
int expectedOffset = 0;
int lastOffset = Int32.MinValue;
foreach (Line line in mLineList) {
// Header lines aren't guaranteed to be sequential and don't have a span.
// They are expected to be in sorted order, and to be unique (with the
// notable exception of the header comment, which is multi-line).
if (line.FileOffset < 0) {
if (line.FileOffset < lastOffset || (line.LineType != Line.Type.LongComment &&
line.FileOffset == lastOffset)) {
Debug.WriteLine("Header offsets went backward: cur=" +
line.FileOffset + " last=" + lastOffset);
return false;
}
lastOffset = line.FileOffset;
continue;
}
// Blank lines and comments can appear before or after code/data. They
// must have the offset of the associated line, and a span of zero.
if (line.FileOffset != expectedOffset && line.FileOffset != lastOffset) {
// arend directives are associated with the last byte of a region, so
// we need to make a special exemption for them.
if (line.LineType == Line.Type.ArEndDirective &&
line.FileOffset == expectedOffset - 1) {
// allow
} else {
Debug.WriteLine("ValidateLineList: bad offset " + line.FileOffset +
" (last=" + lastOffset + ", expected next=" + expectedOffset + ")");
return false;
}
}
if (line.SubLineIndex != 0) {
// In the middle of a multi-line thing, don't advance last/expected.
if (line.LineType != Line.Type.ArEndDirective) {
Debug.Assert(line.FileOffset == lastOffset);
}
} else {
lastOffset = expectedOffset;
expectedOffset += line.OffsetSpan;
}
}
if (expectedOffset != mProject.FileData.Length) {
Debug.WriteLine("ValidateLineList: did not cover entire file: last offset " +
expectedOffset + ", file has " + mProject.FileData.Length);
return false;
}
return true;
}
/// <summary>
/// Generates a synthetic offset for the FileOffset field from an index value.
/// (The exact algorithm isn't too important, as these offsets are not stored in the
/// project file.)
/// </summary>
/// <param name="index">Index into DisasmProject.ActiveDefSymbolListlist.</param>
/// <returns>Synthetic file offset. Value will be < 0.</returns>
public static int DefSymOffsetFromIndex(int index) {
Debug.Assert(index >= 0 && index < (1 << 24));
return index - (1 << 24);
}
/// <summary>
/// Returns the DisasmProject.ActiveDefSymbolList index for an EQU line with
/// the specified file offset.
/// </summary>
/// <param name="offset">Synthetic file offset, from DefSymOffsetFromIndex().</param>
/// <returns>Index into DisasmProject.ActiveDefSymbolListlist.</returns>
public static int DefSymIndexFromOffset(int offset) {
Debug.Assert(offset < 0);
return offset + (1 << 24);
}
// NOTE: the two functions above are tied to the implementation of the function
// below: we output the lines in the order in which they appear in ActiveDefSymbolList.
// If we want to get fancy and sort them, we'll need to do some additional work.
/// <summary>
/// Generates the header lines (header comment, EQU directives).
/// </summary>
/// <param name="proj">Project reference.</param>
/// <param name="formatter">Output formatter.</param>
/// <param name="opNames">Pseudo-op names.</param>
/// <returns>List with header lines.</returns>
private static List<Line> GenerateHeaderLines(DisasmProject proj, Formatter formatter,
PseudoOp.PseudoOpNames opNames) {
List<Line> tmpLines = new List<Line>();
Line line;
// Check for header comment.
if (proj.LongComments.TryGetValue(Line.HEADER_COMMENT_OFFSET,
out MultiLineComment headerComment)) {
List<string> formatted = headerComment.FormatText(formatter, string.Empty);
StringListToLines(formatted, Line.HEADER_COMMENT_OFFSET, Line.Type.LongComment,
CommonWPF.Helper.ZeroColor, 1.0f, tmpLines);
}
// Format symbols.
int index = 0;
foreach (DefSymbol defSym in proj.ActiveDefSymbolList) {
line = new Line(DefSymOffsetFromIndex(index), 0, Line.Type.EquDirective);
// Use an operand length of 1 so things are shown as concisely as possible.
string valueStr = PseudoOp.FormatNumericOperand(formatter, proj.SymbolTable,
null, defSym.DataDescriptor, defSym.Value, 1,
PseudoOp.FormatNumericOpFlags.None);
valueStr = PseudoOp.AnnotateEquDirective(formatter, valueStr, defSym);
string comment = formatter.FormatEolComment(defSym.Comment);
FormattedParts parts = FormattedParts.CreateFullDirective(
defSym.GenerateDisplayLabel(formatter),
formatter.FormatPseudoOp(opNames.EquDirective),
valueStr, comment);
line.Parts = parts;
tmpLines.Add(line);
index++;
}
if (proj.ActiveDefSymbolList.Count != 0) {
// We had some EQUs, throw a blank line at the end.
index++;
line = new Line(DefSymOffsetFromIndex(index), 0, Line.Type.Blank);
tmpLines.Add(line);
}
return tmpLines;
}
/// <summary>
/// Generates lines for the specified range of file offsets.
///
/// Does not generate formatted parts in most cases; that usually happens on demand.
/// Complicated items, such as word-wrapped long comments, may be generated now
/// and saved off.
///
/// This still needs a formatter even when no text is rendered because some
/// options, like maximum per-line operand length, might affect how many lines
/// are generated.
/// </summary>
/// <param name="startOffset">Offset of first byte.</param>
/// <param name="endOffset">Offset of last byte.</param>
/// <param name="lines">List to add output lines to.</param>
private void GenerateLineList(int startOffset, int endOffset, List<Line> lines) {
//Debug.WriteLine("GenerateRange [+" + startOffset.ToString("x6") + ",+" +
// endOffset.ToString("x6") + "]");
Debug.Assert(startOffset >= 0);
Debug.Assert(endOffset >= startOffset);
// Assume variables may have changed.
mLvLookup.Reset();
// Find the previous status flags for M/X tracking. This is for synthesizing
// long/short pseudo-op directives.
StatusFlags prevFlags = StatusFlags.AllIndeterminate;
if (mProject.CpuDef.HasEmuFlag) {
for (int scanoff = startOffset - 1; scanoff >= 0; scanoff--) {
Anattrib attr = mProject.GetAnattrib(scanoff);
if (attr.IsInstructionStart) {
prevFlags = attr.StatusFlags;
// Apply the same tweak here that we do to curFlags below.
prevFlags.M = attr.StatusFlags.IsShortM ? 1 : 0;
prevFlags.X = attr.StatusFlags.IsShortX ? 1 : 0;
Debug.WriteLine("GenerateLineList startOff=+" +
startOffset.ToString("x6") + " using initial flags from +" +
scanoff.ToString("x6") + ": " + prevFlags);
break;
}
}
}
// Configure the initial value of addBlank. The specific case we're handling is
// a no-continue instruction (e.g. JMP) followed by an instruction with a label.
// When we rename the label, we don't want the blank to disappear during the
// partial-list generation.
//
// (Blank lines should always be added ABOVE things rather than AFTER things.)
bool addBlank = false;
if (startOffset > 0) {
int baseOff = DataAnalysis.GetBaseOperandOffset(mProject, startOffset - 1);
if (mProject.GetAnattrib(baseOff).DoesNotContinue) {
// TODO(someday): ideally the blank line would come after inline data
// that follows a no-continue JSR/JSL/BRK
addBlank = true;
}
}
// Create an address map iterator and advance it to match gen.StartOffset.
IEnumerator<AddressMap.AddressChange> addrIter = mProject.AddrMap.AddressChangeIterator;
while (addrIter.MoveNext()) {
if (/*addrIter.Current.IsStart &&*/ addrIter.Current.Offset >= startOffset) {
break;
}
}
bool hasPrgHeader = AsmGen.GenCommon.HasPrgHeader(mProject);
int offset = startOffset;
int arSubLine = 0;
while (offset <= endOffset) {
bool spaceAdded = false;
Anattrib attr = mProject.GetAnattrib(offset);
if (attr.IsInstructionStart && offset > 0 &&
mProject.GetAnattrib(offset - 1).IsData) {
// Transition from data to code. (Don't add blank line for inline data.)
lines.Add(GenerateBlankLine(offset));
spaceAdded = true;
} else if (mProject.VisualizationSets.ContainsKey(offset) && !addBlank) {
// Blank line before visualization set helps keep image visually grouped
// with its data.
lines.Add(GenerateBlankLine(offset));
spaceAdded = true;
} else if (addBlank) {
// Previous instruction wanted to be followed by a blank line.
lines.Add(GenerateBlankLine(offset));
spaceAdded = true;
}
addBlank = false;
// Insert long comments and notes. These may span multiple display lines,
// and require word-wrap, so it's easiest just to render them fully here.
// Set "spaceAdded" to true so arstart doesn't try to add one after the comment.
//
// TODO: integrate into FormattedOperandCache so we don't have to
// regenerate them unless they change. Use the MLC as the dependency.
if (mProject.Notes.TryGetValue(offset, out MultiLineComment noteData)) {
List<string> formatted = noteData.FormatText(mFormatter, "NOTE: ");
StringListToLines(formatted, offset, Line.Type.Note,
noteData.BackgroundColor, NoteColorMultiplier, lines);
spaceAdded = true;
}
if (mProject.LongComments.TryGetValue(offset, out MultiLineComment longComment)) {
List<string> formatted = longComment.FormatText(mFormatter, string.Empty);
StringListToLines(formatted, offset, Line.Type.LongComment,
longComment.BackgroundColor, NoteColorMultiplier, lines);
spaceAdded = true;
}
if (mProject.VisualizationSets.TryGetValue(offset, out VisualizationSet visSet)) {
lines.Add(new Line(offset, 0, Line.Type.VisualizationSet));
spaceAdded = true;
}
// Handle address region starts.
bool multiStart = false;
arSubLine = 0;
while (addrIter.Current != null && addrIter.Current.Offset <= offset) {
AddressMap.AddressChange change = addrIter.Current;
// Range starts/ends shouldn't be embedded in something.
Debug.Assert(change.Offset == offset);
AddressMap.AddressRegion region = change.Region;
if (region.Offset == 0 && hasPrgHeader) {
// Suppress the arstart at offset zero. We know there's another one
// at offset +000002, and that it matches the value at +0/1.
addrIter.MoveNext();
continue;
}
if (change.IsStart) {
// Blank line above ORG directive, except at top of file or when we've
// already added one for another reason.
if (region.Offset != 0 && !spaceAdded && !multiStart) {
lines.Add(GenerateBlankLine(offset));
}
spaceAdded = false; // next one will need a blank line
multiStart = true; // unless it's another arstart immediately
if (region.HasValidPreLabel) {
Line preLine =
new Line(offset, 0, Line.Type.ArStartDirective, arSubLine++);
string preAddrStr = mFormatter.FormatAddress(region.PreLabelAddress,
!mProject.CpuDef.HasAddr16);
preLine.Parts = FormattedParts.CreatePreLabelDirective(preAddrStr,
region.PreLabel);
lines.Add(preLine);
}
Line newLine = new Line(offset, 0, Line.Type.ArStartDirective, arSubLine++);
string addrStr;
if (region.Address == Address.NON_ADDR) {
addrStr = Address.NON_ADDR_STR;
} else if (change.Region.HasValidIsRelative) {
int diff = region.Address - change.Region.PreLabelAddress;
string pfxStr;
if (diff >= 0) {
pfxStr = "*+";
} else {
pfxStr = "*-";
diff = -diff;
}
addrStr = pfxStr + mFormatter.FormatHexValue(diff, 4);
} else {
addrStr = mFormatter.FormatHexValue(region.Address, 4);
}
#if DEBUG1
string comment = mFormatter.FormatEolComment("ends at " +
mFormatter.FormatOffset24(region.Offset + region.ActualLength - 1) +
(region.IsFloating ? " (floating)" : string.Empty));
if (change.IsSynthetic) {
comment += " (auto-generated)";
}
#else
string comment = string.Empty;
if (change.IsSynthetic) {
comment = mFormatter.FormatEolComment("(auto-generated)");
}
#endif
newLine.Parts = FormattedParts.CreateFullDirective(string.Empty,
mFormatter.FormatPseudoOp(mPseudoOpNames.ArStartDirective),
addrStr, comment);
lines.Add(newLine);
addrIter.MoveNext();
} else {
// Next entry is an end marker.
break;
}
}
// Local variable tables come next. Defer rendering.
if (mProject.LvTables.TryGetValue(offset, out LocalVariableTable lvt)) {
int count = lvt.Count;
// If "clear previous" is set, we output an additional line.
if (lvt.ClearPrevious) {
count++;
}
if (count == 0) {
// Need to show something so the user will know an empty table is here.
count = 1;
}
for (int i = 0; i < count; i++) {
Line line = new Line(offset, 0, Line.Type.LocalVariableTable, i);
lines.Add(line);
}
lines.Add(GenerateBlankLine(offset));
}
if (attr.IsInstructionStart) {
// Generate reg width directive, if necessary. These are generated after the
// SEP/REP/PLP instruction has taken effect, so we want it to appear *before*
// the current instruction.
if (mProject.CpuDef.HasEmuFlag) {
// Changing from "ambiguous but assumed short" to "definitively short"
// merits a directive, notably at the start of the file. The tricky
// part is that E=1 means definitively M=1 X=1. And maybe
// indeterminate E also means that.
//
// We don't want to mess with Anattrib, but we do need to tell the
// assembler something. So we tweak our local copy and propagate it.
string operandStr = string.Empty;
StatusFlags curFlags = attr.StatusFlags;
curFlags.M = attr.StatusFlags.IsShortM ? 1 : 0;
curFlags.X = attr.StatusFlags.IsShortX ? 1 : 0;
if (curFlags.M != prevFlags.M) {
operandStr = (curFlags.M == 0) ? "longm" : "shortm";
}
if (curFlags.X != prevFlags.X) {
if (operandStr.Length > 0) {
operandStr += ",";
}
operandStr += (curFlags.X == 0) ? "longx" : "shortx";
}
if (operandStr.Length > 0) {
Line rwLine = new Line(offset, 0, Line.Type.RegWidthDirective);
// FormatPseudoOp isn't quite right for the operand, but there
// isn't anything more suitable, and there are only eight
// possible values. Having the operand capitalization match the
// pseudo-op's feels reasonable.
rwLine.Parts = FormattedParts.CreateDirective(
mFormatter.FormatPseudoOp(mPseudoOpNames.RegWidthDirective),
mFormatter.FormatPseudoOp(operandStr));
lines.Add(rwLine);
}
prevFlags = curFlags;
}
// Add data bank register change. This is generally applied to a PLB
// instruction, so we'd like it to appear on the line following the
// instruction, but that looks funny for e.g. the auto-added B=K at the
// start of each bank, and would be wrong if placed on a "LDA abs" line.
// (We could handle PLB lines differently, but apparently inconsistent
// behavior is unwise.)
if (mProject.DbrChanges.TryGetValue(offset,
out CodeAnalysis.DbrValue dbrValue)) {
string bankStr;
if (dbrValue.FollowPbr) {
bankStr = Res.Strings.DATA_BANK_K;
} else {
bankStr = mFormatter.FormatHexValue(dbrValue.Bank, 2);
}
string fmt;
if (dbrValue.ValueSource == CodeAnalysis.DbrValue.Source.Auto) {
fmt = Res.Strings.DATA_BANK_AUTO_FMT;
} else {
fmt = Res.Strings.DATA_BANK_USER_FMT;
}
string operandStr = string.Format(fmt, bankStr);
Line dbrLine = new Line(offset, 0, Line.Type.DataBankDirective);
dbrLine.Parts = FormattedParts.CreateDirective(
mFormatter.FormatPseudoOp(mPseudoOpNames.DataBankDirective),
mFormatter.FormatPseudoOp(operandStr));
lines.Add(dbrLine);
}
// Look for embedded instructions.
int len;
for (len = 1; len < attr.Length; len++) {
if (mProject.GetAnattrib(offset + len).IsInstructionStart) {
break;
}
}
// Create Line entry. Offset span only covers the instruction up to
// the point where the embedded instruction starts.
Line line = new Line(offset, len, Line.Type.Code);
lines.Add(line);
// Insert blank after an instruction that doesn't continue. Provides a
// break in code, and before a data area.
// TODO(maybe): Might also want to do this if the next offset is data,
// to make things look nicer when code runs directly into data.
//
// We don't want to add it with the current line's offset. If we do that,
// the binary search will get confused, because blank lines have a span
// of zero. If the code is at offset 10 with length 3, and we search for
// the byte at offset 11, then a blank line (with span=0) at offset 10 will
// cause the binary search to assume that the target is farther down, when
// it's actually one line up. We deal with this by setting a flag and
// generating the blank line on the next trip through the loop.
if (attr.DoesNotContinue) {
addBlank = true;
}
offset += len;
} else {
Debug.Assert(attr.DataDescriptor != null);
if (attr.DataDescriptor.IsString) {
// See if we've already got this one.
List<string> strLines = mFormattedLineCache.GetStringEntry(offset,
mFormatter, attr.DataDescriptor, mPseudoOpNames, out string popcode);
if (strLines == null) {
//Debug.WriteLine("FMT string at +" + offset.ToString("x6"));
strLines = PseudoOp.FormatStringOp(mFormatter, mPseudoOpNames,
attr.DataDescriptor, mProject.FileData, offset,
out popcode);
mFormattedLineCache.SetStringEntry(offset, strLines, popcode,
mFormatter, attr.DataDescriptor, mPseudoOpNames);
}
FormattedParts[] partsArray = GenerateStringLines(offset,
popcode, strLines);
for (int i = 0; i < strLines.Count; i++) {
Line line = new Line(offset, attr.Length, Line.Type.Data, i);
line.Parts = partsArray[i];
lines.Add(line);
}
} else {
int numLines = PseudoOp.ComputeRequiredLineCount(mFormatter,
mPseudoOpNames,attr.DataDescriptor, mProject.FileData, offset);
for (int i = 0; i < numLines; i++) {
Line line = new Line(offset, attr.Length, Line.Type.Data, i);
lines.Add(line);
}
}
offset += attr.Length;
}
// Check for address region ends, which will be positioned one byte before the
// updated offset (unless they somehow got embedded inside something else).
//
// The arend can only appear on the same offset as arstart in a single-byte
// region, and we can't have more than one of those at the same offset.
// If this is not a single-byte region, we need to reset the sub-line count.
// (Alternatively: track the offset of the last arstart, and reset subline
// if they differ.)
if (addrIter.Current != null && addrIter.Current.Region.Length != 1) {
arSubLine = 0;
}
while (addrIter.Current != null && addrIter.Current.Offset < offset) {
AddressMap.AddressChange change = addrIter.Current;
// Range starts/ends shouldn't be embedded in something.
if (change.Offset != offset - 1) {
Debug.Assert(false, "Bad offset: change.Offset=+" +
change.Offset.ToString("x6") + " offset-1=+" +
(offset - 1).ToString("x6"));
}
if (change.Region.Offset == 0 && hasPrgHeader) {
// Suppress the arend at offset +000001.
addrIter.MoveNext();
arSubLine++; // need to track address map
continue;
}
if (!change.IsStart) {
// Show the start address to make it easier to pair them visually.
string floatStr = change.Region.IsFloating ? "~" : "";
string addrStr;
if (change.Region.Address == Address.NON_ADDR) {
addrStr = "\u2191 " + floatStr + Address.NON_ADDR_STR;
} else {
addrStr = "\u2191 " + floatStr +
mFormatter.FormatHexValue(change.Region.Address, 4);
}
// Associate with offset of previous byte.
Line newLine = new Line(offset - 1, 0, Line.Type.ArEndDirective,
arSubLine++);
newLine.Parts = FormattedParts.CreateDirective(
mFormatter.FormatPseudoOp(mPseudoOpNames.ArEndDirective),
addrStr);
lines.Add(newLine);
addrIter.MoveNext();
// Put a blank line before the next thing.
// TODO(maybe): this gets lost in a partial update, e.g. you add a
// long comment right after a arend with no following arstart and
// then hit "undo", because the blank is logically part of the
// following offset.
addBlank = true;
} else {
break;
}
}
}
}
/// <summary>
/// Generates a blank line entry.
/// </summary>
private static Line GenerateBlankLine(int offset) {
return new Line(offset, 0, Line.Type.Blank);
}
/// <summary>
/// Takes a list of strings and adds them to the Line list as long comments.
/// </summary>
/// <param name="list">String list.</param>
/// <param name="offset">File offset of item start.</param>
/// <param name="lineType">What type of line this is.</param>
/// <param name="color">Background color (for Notes).</param>
/// <param name="lines">Line list to add data to.</param>
private static void StringListToLines(List<string> list, int offset, Line.Type lineType,
Color color, float mult, List<Line> lines) {
foreach (string str in list) {
Line line = new Line(offset, 0, lineType);
FormattedParts parts = FormattedParts.CreateNote(str,
Color.Multiply(color, mult));
line.Parts = parts;
line.BackgroundColor = color;
lines.Add(line);
}
}
private FormattedParts GenerateInstructionLine(int offset, int instrBytes) {
Anattrib attr = mProject.GetAnattrib(offset);
byte[] data = mProject.FileData;
string offsetStr = mFormatter.FormatOffset24(offset);
int addr = attr.Address;
string addrStr = mFormatter.FormatAddress(addr, !mProject.CpuDef.HasAddr16);
string bytesStr = mFormatter.FormatBytes(data, offset, instrBytes);
string flagsStr = attr.StatusFlags.ToString(mProject.CpuDef.HasEmuFlag);
string attrStr = attr.ToAttrString();
string labelStr = string.Empty;
if (attr.Symbol != null) {
labelStr = attr.Symbol.GenerateDisplayLabel(mFormatter);
}
OpDef op = mProject.CpuDef.GetOpDef(data[offset]);
int operand = op.GetOperand(data, offset, attr.StatusFlags);
int instrLen = op.GetLength(attr.StatusFlags);
OpDef.WidthDisambiguation wdis = OpDef.WidthDisambiguation.None;
if (op.IsWidthPotentiallyAmbiguous) {
wdis = OpDef.GetWidthDisambiguation(instrLen, operand);
}
string opcodeStr = mFormatter.FormatOpcode(op, wdis);
if (attr.Length != instrBytes) {
// An instruction is embedded inside this one. Note that BRK is a two-byte
// instruction, so don't freak out if you see it marked as embedded when a
// $00 is followed by actual code. (But be a little freaked out that your
// code is running into a BRK.)
//opcodeStr = opcodeStr + " \u00bb"; // RIGHT-POINTING DOUBLE ANGLE QUOTATION MARK
//opcodeStr = opcodeStr + " \u23e9"; // BLACK RIGHT-POINTING DOUBLE TRIANGLE
opcodeStr = opcodeStr + " \u25bc"; // BLACK DOWN-POINTING TRIANGLE
}
int operandLen = instrLen - 1;
PseudoOp.FormatNumericOpFlags opFlags = PseudoOp.FormatNumericOpFlags.None;
// Tweak branch instructions. We want to show the absolute address rather
// than the relative offset (which happens with the OperandAddress assignment
// below), and 1-byte branches should always appear as a 4-byte hex value.
if (op.AddrMode == OpDef.AddressMode.PCRel ||
op.AddrMode == OpDef.AddressMode.DPPCRel) {
Debug.Assert(attr.OperandAddress >= 0);
operandLen = 2;
opFlags = PseudoOp.FormatNumericOpFlags.IsPcRel;
} else if (op.AddrMode == OpDef.AddressMode.PCRelLong ||
op.AddrMode == OpDef.AddressMode.StackPCRelLong) {
opFlags = PseudoOp.FormatNumericOpFlags.IsPcRel;
} else if (op.AddrMode == OpDef.AddressMode.Imm ||
op.AddrMode == OpDef.AddressMode.ImmLongA ||
op.AddrMode == OpDef.AddressMode.ImmLongXY) {
opFlags = PseudoOp.FormatNumericOpFlags.HasHashPrefix;
}
if (op.IsAbsolutePBR) {
opFlags |= PseudoOp.FormatNumericOpFlags.IsAbsolutePBR;
}
// Use the OperandAddress when available. This is important for relative branch
// instructions and PER, where we want to show the target address rather than the
// operand value.
int operandForSymbol = operand;
if (attr.OperandAddress >= 0) {
operandForSymbol = attr.OperandAddress;
}
string formattedOperand;
// Check Length to watch for bogus descriptors. ApplyFormatDescriptors() should
// have discarded anything appropriate, so we might be able to eliminate this test.
if (attr.DataDescriptor != null && attr.Length == attr.DataDescriptor.Length) {
Debug.Assert(operandLen > 0);
// Format operand as directed.
if (op.AddrMode == OpDef.AddressMode.BlockMove) {
// Special handling for the double-operand block move.
string opstr1 = PseudoOp.FormatNumericOperand(mFormatter, mProject.SymbolTable,
null, attr.DataDescriptor, operand >> 8, 1,
PseudoOp.FormatNumericOpFlags.None);
string opstr2 = PseudoOp.FormatNumericOperand(mFormatter, mProject.SymbolTable,
null, attr.DataDescriptor, operand & 0xff, 1,
PseudoOp.FormatNumericOpFlags.None);
formattedOperand = '#' + opstr1 + "," + '#' + opstr2;
} else if (op.AddrMode == OpDef.AddressMode.DPPCRel) {
// Special handling for double-operand BBR/BBS. The instruction generally
// behaves like a branch, so format that first.
string branchStr = PseudoOp.FormatNumericOperand(mFormatter,
mProject.SymbolTable, mLvLookup, null, attr.DataDescriptor, offset,
operandForSymbol, operandLen, opFlags);
string dpStr = mFormatter.FormatHexValue(operand & 0xff, 2);
formattedOperand = dpStr + "," + branchStr;
} else {
formattedOperand = PseudoOp.FormatNumericOperand(mFormatter,
mProject.SymbolTable, mLvLookup, null, attr.DataDescriptor, offset,
operandForSymbol, operandLen, opFlags);
}
} else {
// Show operand value in hex.
if (op.AddrMode == OpDef.AddressMode.BlockMove) {
formattedOperand = '#' + mFormatter.FormatHexValue(operand >> 8, 2) + "," +
'#' + mFormatter.FormatHexValue(operand & 0xff, 2);
} else if (op.AddrMode == OpDef.AddressMode.DPPCRel) {
formattedOperand = mFormatter.FormatHexValue(operand & 0xff, 2) + "," +
mFormatter.FormatHexValue(operandForSymbol, operandLen * 2);
} else {
if (operandLen == 2) {
// This is necessary for 16-bit operands, like "LDA abs" and "PEA val",
// when outside bank zero. The bank is included in the operand address,
// but we don't want to show it here.
// (Some assemblers want the bank to be shown for JSR/JMP, but we don't
// do that here. See the corresponding code in AsmGen.GenCommon)
operandForSymbol &= 0xffff;
}
formattedOperand = mFormatter.FormatHexValue(operandForSymbol, operandLen * 2);
}
}
string operandStr = mFormatter.FormatOperand(op, formattedOperand, wdis);
string eolComment = mProject.Comments[offset];
if (mShowCycleCounts) {
bool branchCross = (attr.Address & 0xff00) != (operandForSymbol & 0xff00);
int cycles = mProject.CpuDef.GetCycles(op.Opcode, attr.StatusFlags,
attr.BranchTaken, branchCross);
if (cycles > 0) {
if (!string.IsNullOrEmpty(eolComment)) {
eolComment = cycles.ToString() + " " + eolComment;
} else {
eolComment = cycles.ToString();
}
} else {
if (!string.IsNullOrEmpty(eolComment)) {
eolComment = (-cycles).ToString() + "+ " + eolComment;
} else {
eolComment = (-cycles).ToString() + "+";
}
}
}
string commentStr = mFormatter.FormatEolComment(eolComment);
FormattedParts.PartFlags pflags = FormattedParts.PartFlags.None;
if (mProject.StatusFlagOverrides[offset] != StatusFlags.DefaultValue) {
pflags |= FormattedParts.PartFlags.HasModifiedFlags;
}
if (attr.IsNonAddressable) {
pflags |= FormattedParts.PartFlags.IsNonAddressable;
}
FormattedParts parts = FormattedParts.Create(offsetStr, addrStr, bytesStr,
flagsStr, attrStr, labelStr, opcodeStr, operandStr, commentStr, pflags);
return parts;
}
private FormattedParts GenerateDataLine(int offset, int subLineIndex) {
Anattrib attr = mProject.GetAnattrib(offset);
byte[] data = mProject.FileData;
string offsetStr, addrStr, bytesStr, flagsStr, attrStr, labelStr, opcodeStr,
operandStr, commentStr;
offsetStr = addrStr = bytesStr = flagsStr = attrStr = labelStr = opcodeStr =
operandStr = commentStr = string.Empty;
PseudoOp.PseudoOut pout = PseudoOp.FormatDataOp(mFormatter, mPseudoOpNames,
mProject.SymbolTable, null, attr.DataDescriptor, mProject.FileData, offset,
subLineIndex);
if (subLineIndex == 0) {
offsetStr = mFormatter.FormatOffset24(offset);
addrStr = mFormatter.FormatAddress(attr.Address, !mProject.CpuDef.HasAddr16);
if (attr.Symbol != null) {
labelStr = attr.Symbol.GenerateDisplayLabel(mFormatter);
}
bytesStr = mFormatter.FormatBytes(data, offset, attr.Length);
attrStr = attr.ToAttrString();
opcodeStr = mFormatter.FormatPseudoOp(pout.Opcode);
commentStr = mFormatter.FormatEolComment(mProject.Comments[offset]);
} else {
opcodeStr = " +";
}
operandStr = pout.Operand;
// This seems less useful in practice than in theory, unless you have a lot of
// character data separated by unprintable values. Text generally gets found by
// the data scanner, and having character values next to things that aren't
// actually meant to be character data is more distracting than useful. If
// there's a good use case we could make it an option, and have mCharTest set
// to null if it's switched off.
if (false && attr.DataDescriptor.FormatType == FormatDescriptor.Type.Default) {
FormatDescriptor dfd = attr.DataDescriptor;
Debug.Assert(dfd.Length == 1);
int operand = RawData.GetWord(mProject.FileData, offset, dfd.Length, false);
if (mCharTest((byte)operand)) {
operandStr += " '" + mCharConv((byte)operand) + "'";
}
}
FormattedParts.PartFlags pflags = FormattedParts.PartFlags.None;
if (attr.IsNonAddressable) {
pflags |= FormattedParts.PartFlags.IsNonAddressable;
}
FormattedParts parts = FormattedParts.Create(offsetStr, addrStr, bytesStr,
flagsStr, attrStr, labelStr, opcodeStr, operandStr, commentStr, pflags);
return parts;
}
private FormattedParts GenerateLvTableLine(int offset, int subLineIndex) {
if (!mProject.LvTables.TryGetValue(offset, out LocalVariableTable lvt)) {
Debug.Assert(false);
return FormattedParts.CreateLongComment("BAD OFFSET +" + offset.ToString("x6") +
" sub=" + subLineIndex);
}
if (lvt.ClearPrevious) {
if (subLineIndex == 0) {
return FormattedParts.CreateLongComment(
Res.Strings.LOCAL_VARIABLE_TABLE_CLEAR);
} else {
// adjust for previously output "clear" line
subLineIndex--;
}
}
if (lvt.Count == 0) {
// If ClearPrevious is set, we returned the "clear" line for index zero.
// So this is an empty table without a clear. We want to show something so
// the user knows there's dead weight here.
Debug.Assert(subLineIndex == 0 && !lvt.ClearPrevious);
return FormattedParts.CreateLongComment(
Res.Strings.LOCAL_VARIABLE_TABLE_EMPTY);
}
if (subLineIndex >= lvt.Count) {
return FormattedParts.CreateLongComment("BAD INDEX +" + offset.ToString("x6") +
" sub=" + subLineIndex);
} else {
// Getting the symbol directly from the LvTable yields the original form,
// but we want the de-dup form.
//DefSymbol defSym = lvt[subLineIndex];
List<DefSymbol> lvars = mLvLookup.GetVariablesDefinedAtOffset(offset);
DefSymbol defSym = lvars[subLineIndex];
// Use an operand length of 1 so the value is shown as concisely as possible.
string addrStr = PseudoOp.FormatNumericOperand(mFormatter, mProject.SymbolTable,
null, defSym.DataDescriptor, defSym.Value, 1,
PseudoOp.FormatNumericOpFlags.None);
addrStr = PseudoOp.AnnotateEquDirective(mFormatter, addrStr, defSym);
string comment = mFormatter.FormatEolComment(defSym.Comment);
return FormattedParts.CreateFullDirective(
mFormatter.FormatVariableLabel(defSym.GenerateDisplayLabel(mFormatter)),
mFormatter.FormatPseudoOp(mPseudoOpNames.VarDirective),
addrStr, comment);
}
}
public DefSymbol GetLocalVariableFromLine(int lineIndex) {
Line line = this[lineIndex];
int offset = line.FileOffset;
if (!mProject.LvTables.TryGetValue(offset, out LocalVariableTable lvt)) {
Debug.Assert(false);
return null;
}
int tableIndex = line.SubLineIndex;
if (lvt.ClearPrevious) {
tableIndex--;
}
if (tableIndex < 0 || tableIndex >= lvt.Count) {
// Will be -1 on first line when ClearPrevious was set. Will be zero on
// first line of empty table.
return null;
}
// Go through LvLookup to get de-dup form of labels.
//return lvt[tableIndex];
List<DefSymbol> lvars = mLvLookup.GetVariablesDefinedAtOffset(offset);
return lvars[tableIndex];
}
public AddressMap.AddressRegion GetAddrRegionFromLine(Line line, out bool isSynth) {
// A given offset can have one or more arstart lines and one or more arend lines.
// You can't have an end followed by a start, because that would mean the regions
// overlap. If there's both start and end present, we have a 1-byte region.
int offset = line.FileOffset;
List<AddressMap.AddressMapEntry> entries = mProject.AddrMap.GetEntries(offset);
IEnumerator<AddressMap.AddressChange> addrIter = mProject.AddrMap.AddressChangeIterator;
while (addrIter.MoveNext()) {
if (addrIter.Current.Offset >= offset) {
break;
}
}
int count = line.SubLineIndex;
while (addrIter.Current != null && addrIter.Current.Offset == offset) {
AddressMap.AddressChange change = addrIter.Current;
if (count == 0) {
// Could be pre-label or arstart that follows it. Both go to the same place.
isSynth = change.IsSynthetic;
return change.Region;
}
if (change.IsStart && change.Region.HasValidPreLabel) {
count--;
if (count == 0) {
// This is the arstart following the pre-label.
isSynth = change.IsSynthetic;
return change.Region;
}
}
count--;
addrIter.MoveNext();
}
isSynth = false;
return null;
}
private FormattedParts[] GenerateStringLines(int offset, string popcode,
List<string> operands) {
FormattedParts[] partsArray = new FormattedParts[operands.Count];
Anattrib attr = mProject.GetAnattrib(offset);
byte[] data = mProject.FileData;
string offsetStr, addrStr, bytesStr, attrStr, labelStr, opcodeStr,
operandStr, commentStr;
for (int subLineIndex = 0; subLineIndex < operands.Count; subLineIndex++) {
if (subLineIndex == 0) {
offsetStr = mFormatter.FormatOffset24(offset);
addrStr = mFormatter.FormatAddress(attr.Address, !mProject.CpuDef.HasAddr16);
if (attr.Symbol != null) {
labelStr = attr.Symbol.GenerateDisplayLabel(mFormatter);
} else {
labelStr = string.Empty;
}
bytesStr = mFormatter.FormatBytes(data, offset, attr.Length);
attrStr = attr.ToAttrString();
opcodeStr = mFormatter.FormatPseudoOp(popcode);
commentStr = mFormatter.FormatEolComment(mProject.Comments[offset]);
} else {
offsetStr = addrStr = bytesStr = attrStr = labelStr = commentStr =
string.Empty;
opcodeStr = " +";
}
operandStr = operands[subLineIndex];
FormattedParts.PartFlags pflags = FormattedParts.PartFlags.None;
if (attr.IsNonAddressable) {
pflags |= FormattedParts.PartFlags.IsNonAddressable;
}
FormattedParts parts = FormattedParts.Create(offsetStr, addrStr, bytesStr,
/*flags*/string.Empty, attrStr, labelStr, opcodeStr, operandStr, commentStr,
pflags);
partsArray[subLineIndex] = parts;
}
return partsArray;
}
}
}