From 29a96d3c098cefe7cb388ee914b8e496d7970406 Mon Sep 17 00:00:00 2001
From: Vadim Zeitlin <vadim@wxwidgets.org>
Date: Sun, 11 May 2008 17:44:32 +0000
Subject: [PATCH] avoid extra reallocations and memcpy for seekable files in
OnRead() (slightly modified patch 1935975)
git-svn-id: https://svn.wxwidgets.org/svn/wx/wxWidgets/trunk@53557 c3d73ce0-8a6f-49c7-b76d-6d57e0e08775
---
src/common/textfile.cpp | 98 ++++++++++++++++++++++++-----------------
1 file changed, 57 insertions(+), 41 deletions(-)
diff --git a/src/common/textfile.cpp b/src/common/textfile.cpp
index ef27836e89..db0d4b41ba 100644
--- a/src/common/textfile.cpp
+++ b/src/common/textfile.cpp
@@ -92,15 +92,17 @@ bool wxTextFile::OnRead(const wxMBConv& conv)
wxASSERT_MSG( m_file.IsOpened(), _T("can't read closed file") );
// read the entire file in memory: this is not the most efficient thing to
- // do but there is no good way to avoid it in Unicode build because if we
- // read the file block by block we can't convert each block to Unicode
+ // do it but there is no good way to avoid it in Unicode build because if
+ // we read the file block by block we can't convert each block to Unicode
// separately (the last multibyte char in the block might be only partially
// read and so the conversion would fail) and, as the file contents is kept
// in memory by wxTextFile anyhow, it shouldn't be a big problem to read
// the file entirely
- size_t bufSize = 0,
- bufPos = 0;
- char block[1024];
+ size_t bufSize = 0;
+
+ // number of bytes to (try to) read from disk at once
+ static const size_t BLOCK_SIZE = 4096;
+
wxCharBuffer buf;
// first determine if the file is seekable or not and so whether we can
@@ -118,57 +120,70 @@ bool wxTextFile::OnRead(const wxMBConv& conv)
{
// we know the required length, so set the buffer size in advance
bufSize = fileLength;
- if ( !buf.extend(bufSize - 1 /* it adds 1 internally */) )
+ if ( !buf.extend(bufSize) )
return false;
// if the file is seekable, also check that we're at its beginning
wxASSERT_MSG( m_file.Tell() == 0, _T("should be at start of file") );
- }
- for ( ;; )
- {
- ssize_t nRead = m_file.Read(block, WXSIZEOF(block));
-
- if ( nRead == wxInvalidOffset )
+ char *dst = buf.data();
+ for ( ;; )
{
- // read error (error message already given in wxFile::Read)
- return false;
- }
+ ssize_t nRead = m_file.Read(dst, BLOCK_SIZE);
- if ( nRead == 0 )
- {
- // if no bytes have been read, presumably this is a valid-but-empty file
- if ( bufPos == 0 )
- return true;
+ if ( nRead == wxInvalidOffset )
+ {
+ // read error (error message already given in wxFile::Read)
+ return false;
+ }
- // otherwise we've finished reading the file
- break;
- }
+ if ( nRead == 0 )
+ {
+ // this file can't be empty because we checked for this above
+ break;
+ }
- if ( seekable )
- {
- // this shouldn't happen but don't overwrite the buffer if it does
- wxCHECK_MSG( bufPos + nRead <= bufSize, false,
- _T("read more than file length?") );
+ dst += nRead;
}
- else // !seekable
+
+ wxASSERT_MSG( dst - buf.data() == (wxFileOffset)bufSize,
+ _T("logic error") );
+ }
+ else // file is not seekable
+ {
+ char block[BLOCK_SIZE];
+ for ( ;; )
{
- // for non-seekable files we have to allocate more memory on the go
- if ( !buf.extend(bufPos + nRead - 1 /* it adds 1 internally */) )
+ ssize_t nRead = m_file.Read(block, WXSIZEOF(block));
+
+ if ( nRead == wxInvalidOffset )
+ {
+ // read error (error message already given in wxFile::Read)
return false;
- }
+ }
- // append to the buffer
- memcpy(buf.data() + bufPos, block, nRead);
- bufPos += nRead;
- }
+ if ( nRead == 0 )
+ {
+ // if no bytes have been read, presumably this is a
+ // valid-but-empty file
+ if ( bufSize == 0 )
+ return true;
- if ( !seekable )
- {
- bufSize = bufPos;
+ // otherwise we've finished reading the file
+ break;
+ }
+
+ // extend the buffer for new data
+ if ( !buf.extend(bufSize + nRead) )
+ return false;
+
+ // and append it to the buffer
+ memcpy(buf.data() + bufSize, block, nRead);
+ bufSize += nRead;
+ }
}
- const wxString str(buf, conv, bufPos);
+ const wxString str(buf, conv, bufSize);
// there's no risk of this happening in ANSI build
#if wxUSE_UNICODE
@@ -179,7 +194,8 @@ bool wxTextFile::OnRead(const wxMBConv& conv)
}
#endif // wxUSE_UNICODE
- free(buf.release()); // we don't need this memory any more
+ // we don't need this memory any more
+ buf.reset();
// now break the buffer in lines
--
2.47.2