[glom] Refactoring Glom::Dialog_Import_CSV
- From: Murray Cumming <murrayc src gnome org>
- To: svn-commits-list gnome org
- Cc:
- Subject: [glom] Refactoring Glom::Dialog_Import_CSV
- Date: Thu, 10 Sep 2009 08:43:03 +0000 (UTC)
commit 8076d24b2c38e8824270d70f56ed15422be881a9
Author: Michael Hasselmann <michaelh openismus com>
Date: Fri Sep 4 11:11:40 2009 +0200
Refactoring Glom::Dialog_Import_CSV
* ChangeLog, Makefile_glom.am, glom/dialog_import_csv.[h|cc],
glom/import_csv.[h|cc], glom/dialog_import_csv_progress.cc: Moved the inlined
Parser class from the Dialog_Import_CSV class to the new CsvParser class. The
CsvParser already handles the line scanning and encoding conversion but is
supposed to take over the file reading responsibility as well. In order to
allow the Dialog_Import_CSV class to parse a scanned line into its tree model
the parser emits a lineScanned signal, with the currently scanned line as
parameter.
ChangeLog | 13 +
Makefile_glom.am | 2 +
glom/dialog_import_csv.cc | 486 +++++++++++-------------------------
glom/dialog_import_csv.h | 63 ++----
glom/dialog_import_csv_progress.cc | 10 +-
glom/import_csv.cc | 314 +++++++++++++++++++++++
glom/import_csv.h | 104 ++++++++
7 files changed, 605 insertions(+), 387 deletions(-)
---
diff --git a/ChangeLog b/ChangeLog
index 8625acb..9571b2d 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,16 @@
+2009-09-09 Michael Hasselmann <michaelh openismus com>
+
+ Refactoring Glom::Dialog_Import_CSV
+
+ * ChangeLog, Makefile_glom.am, glom/dialog_import_csv.[h|cc],
+ glom/import_csv.[h|cc], glom/dialog_import_csv_progress.cc: Moved the inlined
+ Parser class from the Dialog_Import_CSV class to the new CsvParser class. The
+ CsvParser already handles the line scanning and encoding conversion but is
+ supposed to take over the file reading responsibility as well. In order to
+ allow the Dialog_Import_CSV class to parse a scanned line into its tree model
+ the parser emits a lineScanned signal, with the currently scanned line as
+ parameter.
+
2009-09-08 Johannes Schmid <jschmid openismus com>
Bug 594357 - Maemo: sqlite behaves strange
diff --git a/Makefile_glom.am b/Makefile_glom.am
index a619535..5057657 100644
--- a/Makefile_glom.am
+++ b/Makefile_glom.am
@@ -54,6 +54,8 @@ glom_glom_SOURCES = \
glom/glom_postgres.h \
glom/glom_privs.cc \
glom/glom_privs.h \
+ glom/import_csv.cc \
+ glom/import_csv.h \
glom/main.cc \
glom/notebook_glom.cc \
glom/notebook_glom.h \
diff --git a/glom/dialog_import_csv.cc b/glom/dialog_import_csv.cc
index e53f2df..8dc616a 100644
--- a/glom/dialog_import_csv.cc
+++ b/glom/dialog_import_csv.cc
@@ -29,98 +29,9 @@
#include <glibmm/i18n.h>
#include <cerrno>
-// On Windows, "iconv" seems to be a define for "libiconv", breaking the Glib::IConv::iconv() call.
-#ifdef iconv
-#undef iconv
-#endif
-
namespace
{
-const gunichar DELIMITER = ',';
-const gunichar QUOTE = '\"';
-
-static bool next_char_is_quote(const Glib::ustring::const_iterator& iter, const Glib::ustring::const_iterator& end)
-{
- if(iter == end)
- return false;
-
- //Look at the next character to see if it's really "" (an escaped "):
- Glib::ustring::const_iterator iter_next = iter;
- ++iter_next;
- if(iter_next != end)
- {
- const gunichar c_next = *iter_next;
- if(c_next == QUOTE)
- {
- return true;
- }
- }
-
- return false;
-}
-
-//Parse the field in a comma-separated line, returning the field including the quotes:
-static Glib::ustring::const_iterator advance_field(const Glib::ustring::const_iterator& iter, const Glib::ustring::const_iterator& end, Glib::ustring& field)
-{
- bool inside_quotes = false;
- //bool string_finished = false; //Ignore anything after "something", such as "something"else,
-
- field.clear();
-
- Glib::ustring::const_iterator walk;
- for(walk = iter; walk != end; ++walk)
- {
- const gunichar c = *walk;
-
- //if(string_finished)
- // continue;
-
- if(inside_quotes)
- {
- // End of quoted string?
- if(c == QUOTE)
- {
- if(next_char_is_quote(walk, end))
- {
- //This is "" so it's not an end quote. Just add one quote:
- field += c;
- ++walk; //Skip the second ".
- }
- else
- {
- inside_quotes = false;
- //string_finished = true; //Ignore anything else before the next comma.
- }
-
- continue;
- }
- }
- else
- {
- // Start of quoted string:
- if((c == QUOTE))
- {
- inside_quotes = true;
- continue;
- }
- // End of field:
- else if(!inside_quotes && c == DELIMITER)
- {
- break;
- }
-
- continue;
- }
-
- field += c; // Just so that we don't need to iterate through the field again, since there is no Glib::ustring::substr(iter, iter)
- }
-
- // TODO: Throw error if still inside a quoted string?
- std::cout << "debug: field=" << field << std::endl;
- return walk;
-}
-
struct Encoding {
const char* name;
const char* charset;
@@ -192,7 +103,8 @@ namespace Glom
{
Dialog_Import_CSV::Dialog_Import_CSV(BaseObjectType* cobject, const Glib::RefPtr<Gtk::Builder>& builder)
-: Gtk::Dialog(cobject), m_auto_detect_encoding(0), m_state(NONE)
+: Gtk::Dialog(cobject),
+ m_auto_detect_encoding()
{
builder->get_widget("import_csv_fields", m_sample_view);
builder->get_widget("import_csv_target_table", m_target_table);
@@ -231,7 +143,14 @@ Dialog_Import_CSV::Dialog_Import_CSV(BaseObjectType* cobject, const Glib::RefPtr
m_encoding_combo->set_cell_data_func(*renderer, sigc::bind(sigc::mem_fun(*this, &Dialog_Import_CSV::encoding_data_func), sigc::ref(*renderer)));
m_encoding_combo->set_row_separator_func(sigc::mem_fun(*this, &Dialog_Import_CSV::row_separator_func));
m_encoding_combo->set_active(0);
+
m_encoding_combo->signal_changed().connect(sigc::mem_fun(*this, &Dialog_Import_CSV::on_encoding_changed));
+
+ // TODO: Reset parser encoding on selection changed.
+ m_parser = std::auto_ptr<CsvParser>(new CsvParser(get_current_encoding().c_str()));
+ m_parser->signal_encoding_error().connect(sigc::mem_fun(*this, &Dialog_Import_CSV::on_encoding_error));
+ m_parser->signal_line_scanned().connect(sigc::mem_fun(*this, &Dialog_Import_CSV::on_line_scanned));
+
m_first_line_as_title->signal_toggled().connect(sigc::mem_fun(*this, &Dialog_Import_CSV::on_first_line_as_title_toggled));
m_sample_rows->signal_changed().connect(sigc::mem_fun(*this, &Dialog_Import_CSV::on_sample_rows_changed));
@@ -258,6 +177,21 @@ Dialog_Import_CSV::Dialog_Import_CSV(BaseObjectType* cobject, const Glib::RefPtr
clear();
}
+CsvParser::State Dialog_Import_CSV::get_parser_state() const
+{
+ return m_parser->m_state;
+}
+
+Glib::ustring Dialog_Import_CSV::get_target_table_name() const
+{
+ return m_target_table->get_text();
+}
+
+const Glib::ustring& Dialog_Import_CSV::get_filename() const
+{
+ return m_filename;
+}
+
void Dialog_Import_CSV::import(const Glib::ustring& uri, const Glib::ustring& into_table)
{
clear();
@@ -313,16 +247,16 @@ void Dialog_Import_CSV::import(const Glib::ustring& uri, const Glib::ustring& in
// Query the display name of the file to set in the title:
m_file->query_info_async(sigc::mem_fun(*this, &Dialog_Import_CSV::on_query_info), G_FILE_ATTRIBUTE_STANDARD_DISPLAY_NAME);
- set_state(PARSING);
+ set_parser_state(CsvParser::PARSING);
}
}
guint Dialog_Import_CSV::get_row_count() const
{
- if(m_first_line_as_title->get_active() && m_rows.size() > 1)
- return m_rows.size() - 1;
+ if(m_first_line_as_title->get_active() && m_parser->m_rows.size() > 1)
+ return m_parser->m_rows.size() - 1;
- return m_rows.size();
+ return m_parser->m_rows.size();
}
guint Dialog_Import_CSV::get_column_count() const
@@ -340,7 +274,7 @@ const Glib::ustring& Dialog_Import_CSV::get_data(guint row, guint col)
if(m_first_line_as_title->get_active())
++row;
- return m_rows[row][col];
+ return m_parser->m_rows[row][col];
}
void Dialog_Import_CSV::clear()
@@ -352,22 +286,19 @@ void Dialog_Import_CSV::clear()
m_sample_view->set_model(m_sample_model);
m_field_model.reset();
m_field_model_sorted.reset();
- m_rows.clear();
m_fields.clear();
m_file.reset();
- m_stream.reset();
m_filename.clear();
m_buffer.reset(NULL);
- m_raw.clear();
- m_parser.reset(NULL);
-
+ m_parser->clear();
+ //m_parser.reset(NULL);
m_encoding_info->set_text("");
m_sample_view->set_sensitive(false);
m_encoding_combo->set_sensitive(false);
m_first_line_as_title->set_sensitive(false);
m_sample_rows->set_sensitive(false);
- set_state(NONE);
+ set_parser_state(CsvParser::NONE);
validate_primary_key();
}
@@ -421,10 +352,10 @@ void Dialog_Import_CSV::on_file_read(const Glib::RefPtr<Gio::AsyncResult>& resul
#ifdef GLIBMM_EXCEPTIONS_ENABLED
try
{
- m_stream = m_file->read_finish(result);
+ m_parser->m_stream = m_file->read_finish(result);
m_buffer.reset(new Buffer);
- m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
+ m_parser->m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
}
catch(const Glib::Exception& error)
{
@@ -434,11 +365,11 @@ void Dialog_Import_CSV::on_file_read(const Glib::RefPtr<Gio::AsyncResult>& resul
}
#else
std::auto_ptr<Glib::Error> error;
- m_stream = m_file->read_finish(result, error);
+ m_parser->m_stream = m_file->read_finish(result, error);
if (!error.get())
{
m_buffer.reset(new Buffer);
- m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
+ m_parser->m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
}
else
{
@@ -453,19 +384,20 @@ void Dialog_Import_CSV::on_stream_read(const Glib::RefPtr<Gio::AsyncResult>& res
#ifdef GLIBMM_EXCEPTIONS_ENABLED
try
{
- const gssize size = m_stream->read_finish(result);
- m_raw.insert(m_raw.end(), m_buffer->buf, m_buffer->buf + size);
+ const gssize size = m_parser->m_stream->read_finish(result);
+ m_parser->m_raw.insert(m_parser->m_raw.end(), m_buffer->buf, m_buffer->buf + size);
// If the parser already exists, but it is currently not parsing because it waits
// for new input, then continue parsing.
- if(m_parser.get() && !m_parser->idle_connection.connected())
+ // TODO: Introduce CsvParser::is_idle_handler_connected() instead?
+ if(m_parser.get() && !m_parser->m_idle_connection.connected())
{
- m_parser->idle_connection = Glib::signal_idle().connect(sigc::mem_fun(*this, &Dialog_Import_CSV::on_idle_parse));
+ m_parser->m_idle_connection = Glib::signal_idle().connect(sigc::mem_fun(*m_parser.get(), &CsvParser::on_idle_parse));
}
// If the parser does not exist yet, then create a new parser, except when the
// current encoding does not work for the file ,in which case the user must first
// choose another encoding.
- else if(!m_parser.get() && m_state != ENCODING_ERROR)
+ else if(!m_parser.get() && get_parser_state() != CsvParser::ENCODING_ERROR)
{
begin_parse();
}
@@ -473,13 +405,13 @@ void Dialog_Import_CSV::on_stream_read(const Glib::RefPtr<Gio::AsyncResult>& res
if(size > 0)
{
// Read the next few bytes
- m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
+ m_parser->m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
}
else
{
// Finished reading
m_buffer.reset(NULL);
- m_stream.reset();
+ m_parser->m_stream.reset();
m_file.reset();
}
}
@@ -491,10 +423,10 @@ void Dialog_Import_CSV::on_stream_read(const Glib::RefPtr<Gio::AsyncResult>& res
}
#else
std::auto_ptr<Glib::Error> error;
- const gssize size = m_stream->read_finish(result, error);
+ const gssize size = m_parser->m_stream->read_finish(result, error);
if (!error.get())
{
- m_raw.insert(m_raw.end(), m_buffer->buf, m_buffer->buf + size);
+ m_parser->m_raw.insert(m_parser->m_raw.end(), m_buffer->buf, m_buffer->buf + size);
// If the parser already exists, but it is currently not parsing because it waits
// for new input, then continue parsing.
@@ -505,7 +437,7 @@ void Dialog_Import_CSV::on_stream_read(const Glib::RefPtr<Gio::AsyncResult>& res
// If the parser does not exist yet, then create a new parser, except when the
// current encoding does not work for the file ,in which case the user must first
// choose another encoding.
- else if(!m_parser.get() && m_state != ENCODING_ERROR)
+ else if(!m_parser.get() && get_parser_state() != CsvParser::ENCODING_ERROR)
{
begin_parse();
}
@@ -513,13 +445,13 @@ void Dialog_Import_CSV::on_stream_read(const Glib::RefPtr<Gio::AsyncResult>& res
if(size > 0)
{
// Read the next few bytes
- m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
+ m_parser->m_stream->read_async(m_buffer->buf, sizeof(m_buffer->buf), sigc::mem_fun(*this, &Dialog_Import_CSV::on_stream_read));
}
else
{
// Finished reading
m_buffer.reset(NULL);
- m_stream.reset();
+ m_parser->m_stream.reset();
m_file.reset();
}
}
@@ -533,9 +465,6 @@ void Dialog_Import_CSV::on_stream_read(const Glib::RefPtr<Gio::AsyncResult>& res
void Dialog_Import_CSV::on_encoding_changed()
{
- // Reset current parsing process
- m_parser.reset(NULL);
-
int active = m_encoding_combo->get_active_row_number();
switch(active)
{
@@ -551,9 +480,15 @@ void Dialog_Import_CSV::on_encoding_changed()
break;
}
+ // Reset current parsing process
+ // TODO: Troublesome. Parser now contains a bit more members, not sure we can simply reset it like that.
+ //m_parser.reset(NULL);
+ //m_parser->clear();
+ //m_parser->set_encoding(get_current_encoding().c_str());
+
// Parse from beginning with new encoding if we have already some data to
// parse.
- if(!m_raw.empty())
+ if(!m_parser->m_raw.empty())
begin_parse();
}
@@ -577,7 +512,7 @@ void Dialog_Import_CSV::on_first_line_as_title_toggled()
// Add another row to the end, if one is loaded.
const guint last_index = m_sample_model->children().size();
- if(last_index < m_rows.size())
+ if(last_index < m_parser->m_rows.size())
{
iter = m_sample_model->append();
(*iter)[m_sample_columns.m_col_row] = last_index;
@@ -592,7 +527,7 @@ void Dialog_Import_CSV::on_first_line_as_title_toggled()
Gtk::TreeModel::Path path("1");
Gtk::TreeModel::iterator iter = m_sample_model->get_iter(path);
- if((!iter || (*iter)[m_sample_columns.m_col_row] != 0) && !m_rows.empty() && m_sample_rows->get_value_as_int() > 0)
+ if((!iter || (*iter)[m_sample_columns.m_col_row] != 0) && !m_parser->m_rows.empty() && m_sample_rows->get_value_as_int() > 0)
{
// Add first row to model
if(!iter)
@@ -642,7 +577,7 @@ void Dialog_Import_CSV::on_sample_rows_changed()
if(m_first_line_as_title->get_active())
++row_index;
- for(guint i = current_sample_rows; i < new_sample_rows && row_index < m_rows.size(); ++i, ++row_index)
+ for(guint i = current_sample_rows; i < new_sample_rows && row_index < m_parser->m_rows.size(); ++i, ++row_index)
{
Gtk::TreeModel::iterator iter = m_sample_model->append();
(*iter)[m_sample_columns.m_col_row] = row_index;
@@ -662,6 +597,7 @@ Glib::ustring Dialog_Import_CSV::get_current_encoding() const
return AUTODETECT_ENCODINGS[m_auto_detect_encoding].charset;
}
+ // TODO: change return type?
return encoding.c_str();
}
@@ -677,29 +613,29 @@ void Dialog_Import_CSV::begin_parse()
m_sample_model.reset();
m_sample_view->remove_all_columns();
m_sample_view->set_model(m_sample_model); // Empty model
- m_rows.clear();
+ m_parser->clear();
- m_parser.reset(new Parser(get_current_encoding().c_str()));
- set_state(PARSING);
+ m_parser->set_encoding(get_current_encoding().c_str());
+ set_parser_state(CsvParser::PARSING);
// Allow the Import button to be pressed when a field for the primary key
// field is set. When the import button is pressed without the file being
// fully loaded, the import progress waits for us to load the rest.
validate_primary_key();
- m_parser->idle_connection = Glib::signal_idle().connect(sigc::mem_fun(*this, &Dialog_Import_CSV::on_idle_parse));
+ m_parser->m_idle_connection = Glib::signal_idle().connect(sigc::mem_fun(*m_parser.get(), &CsvParser::on_idle_parse));
}
-void Dialog_Import_CSV::encoding_error()
+void Dialog_Import_CSV::on_encoding_error()
{
- m_parser.reset(NULL);
+ m_parser->clear();
// Clear sample preview (TODO: Let it visible, and only remove when reparsing?)
m_sample_model.reset();
m_sample_view->remove_all_columns();
m_sample_view->set_model(m_sample_model); // Empty model
- m_rows.clear();
- set_state(ENCODING_ERROR);
+ // TODO: move into parser.
+ set_parser_state(CsvParser::ENCODING_ERROR);
// Don't allow the import button to be pressed when an error occured. This
// would not make sense since we cleared all the parsed row data anyway.
@@ -720,216 +656,42 @@ void Dialog_Import_CSV::encoding_error()
}
}
-bool Dialog_Import_CSV::on_idle_parse()
-{
- // The amount of bytes to process in one pass of the idle handler:
- static const guint CONVERT_BUFFER_SIZE = 1024;
-
- const char* inbuffer = &m_raw[m_parser->input_position];
- char* inbuf = const_cast<char*>(inbuffer);
- gsize inbytes = m_raw.size() - m_parser->input_position;
- char outbuffer[CONVERT_BUFFER_SIZE];
- char* outbuf = outbuffer;
- gsize outbytes = CONVERT_BUFFER_SIZE;
-
- const std::size_t result = m_parser->conv.iconv(&inbuf, &inbytes, &outbuf, &outbytes);
- bool more_to_process = (inbytes != 0);
-
- if(result == static_cast<size_t>(-1))
- {
- if(errno == EILSEQ)
- {
- // Invalid text in the current encoding.
- encoding_error();
- return false;
- }
-
- // If EINVAL is set, this means that an incomplete multibyte sequence was at
- // the end of the input. We might have some more bytes, but those do not make
- // up a whole character, so we need to wait for more input.
- if(errno == EINVAL)
- {
- if(!m_stream)
- {
- // This means that we already reached the end of the file. The file
- // should not end with an incomplete multibyte sequence.
- encoding_error();
- return false;
- }
- else
- {
- more_to_process = false;
- }
- }
- }
-
- m_parser->input_position += (inbuf - inbuffer);
-
- // We now have outbuf - outbuffer bytes of valid UTF-8 in outbuffer.
- const char* prev_line_end = outbuffer;
- const char* prev = prev_line_end;
-
- //Identify the record rows in the .cvs file.
- //We can't just search for newlines because they may be inside quotes too.
- //TODO: Use a regex instead, to more easily handle quotes?
- bool in_quotes = false;
- while(true)
- {
- //Note that, unlike std::string::find*, std::find* returns an iterator (char*), not a position.
- //It returns outbuf if none is found.
- const char newline_to_find[] = { '\r', '\n', '\0' };
- const char* pos_newline = std::find_first_of<const char*>(prev, outbuf, newline_to_find, newline_to_find + sizeof(newline_to_find));
-
- const char quote_to_find[] = {(char)QUOTE};
- const char* pos_quote = std::find_first_of<const char*>(prev, outbuf, quote_to_find, quote_to_find + sizeof(quote_to_find));
-
- //Examine the first character (quote or newline) that was found:
- const char* pos = pos_newline;
- if((pos_quote != outbuf) && pos_quote < pos)
- pos = pos_quote;
-
- if(pos == outbuf)
- break;
-
- char ch = *pos;
-
- if(ch == '\0')
- {
- // There is a null byte in the conversion. Because normal text files don't
- // contain null bytes this only occurs when converting, for example, a UTF-16
- // file from ISO-8859-1 to UTF-8 (note that the UTF-16 file is valid ISO-8859-1 -
- // it just contains lots of nullbytes). We therefore produce an error here.
- encoding_error();
- return false;
- }
- else if(in_quotes)
- {
- //Ignore newlines inside quotes.
-
- //End quote:
- if(ch == (char)QUOTE)
- in_quotes = false;
-
- prev = pos + 1;
- continue;
- }
- else
- {
- //Start quote:
- if(ch == (char)QUOTE)
- {
- in_quotes = true;
- prev = pos + 1;
- continue;
- }
-
- //Found a newline (outside of quotes) that marks the end of the line:
- m_parser->current_line.append(prev_line_end, pos - prev_line_end);
- ++(m_parser->line_number);
-
- if(!m_parser->current_line.empty())
- handle_line(m_parser->current_line, m_parser->line_number);
-
- m_parser->current_line.clear();
-
- // Skip linebreak
- prev = pos + 1;
-
- // Skip DOS-style linebreak (\r\n)
- if(ch == '\r'
- && prev != outbuf && *prev == '\n')
- {
- ++prev;
- }
-
- prev_line_end = prev;
- }
- }
-
- // Append last chunk of this line
- m_parser->current_line.append(prev, outbuf - prev);
- if(!m_stream && m_raw.size() == m_parser->input_position)
- {
- ++(m_parser->line_number);
-
- // Handle last line, if nonempty
- if(m_parser->current_line.empty())
- handle_line(m_parser->current_line, m_parser->line_number);
-
- // Parsed whole file, done
- m_parser.reset(NULL);
- set_state(PARSED);
- }
-
- // Continue if there are more bytes to process
- return more_to_process;
-}
-
-void Dialog_Import_CSV::handle_line(const Glib::ustring& line, guint line_number)
+/*
+ * No, this is wrong. Creating the tree model and handling a line from the CSV file are two separate steps. Proposal: Construct tree model *after* parsing, using row[0].
+ */
+void Dialog_Import_CSV::on_line_scanned(const Glib::ustring& line, guint line_number)
{
- std::cout << "debug: handle_line=" << line << std::endl;
+ std::cout << "debug: on_line_scanned=" << line << std::endl;
if(line.empty())
return;
- m_rows.push_back(type_row_strings());
- type_row_strings& row = m_rows.back();
+ m_parser->m_rows.push_back(CsvParser::type_row_strings());
+ CsvParser::type_row_strings& row = m_parser->m_rows.back();
Glib::ustring field;
//Gtk::TreeModelColumnRecord record;
// Parse first field:
- Glib::ustring::const_iterator iter = advance_field(line.begin(), line.end(), field);
+ Glib::ustring::const_iterator line_iter = CsvParser::advance_field(line.begin(), line.end(), field);
row.push_back(field);
// Parse more fields:
- while(iter != line.end())
+ while(line_iter != line.end())
{
// Skip delimiter:
- ++ iter;
+ ++line_iter;
// Read field:
- iter = advance_field(iter, line.end(), field);
+ line_iter = CsvParser::advance_field(line_iter, line.end(), field);
// Add field to current row:
row.push_back(field);
}
+ // This is the first line read if there is no model yet:
if(!m_sample_model)
{
- // This is the first line read if there is no model yet:
- m_sample_model = Gtk::ListStore::create(m_sample_columns);
- m_sample_view->set_model(m_sample_model);
-
- // Create field vector that contains the fields into which to import
- // the data.
- m_fields.resize(row.size());
-
- Gtk::CellRendererText* text = Gtk::manage(new Gtk::CellRendererText);
- Gtk::TreeViewColumn* col = Gtk::manage(new Gtk::TreeViewColumn(_("Line")));
- col->pack_start(*text, false);
- col->set_cell_data_func(*text, sigc::mem_fun(*this, &Dialog_Import_CSV::line_data_func));
- m_sample_view->append_column(*col);
-
- for(guint i = 0; i < row.size(); ++ i)
- {
- Gtk::CellRendererCombo* cell = Gtk::manage(new Gtk::CellRendererCombo);
- Gtk::TreeViewColumn* col = Gtk::manage(new Gtk::TreeViewColumn(row[i]));
- col->pack_start(*cell, true);
- col->set_cell_data_func(*cell, sigc::bind(sigc::mem_fun(*this, &Dialog_Import_CSV::field_data_func), i));
- col->set_sizing(Gtk::TREE_VIEW_COLUMN_AUTOSIZE);
-#ifdef GLIBMM_PROPERTIES_ENABLED
- cell->property_model() = m_field_model_sorted;
- cell->property_text_column() = 0;
- cell->property_has_entry() = false;
-#else
- cell->set_property("model", m_field_model_sorted);
- cell->set_property("text-column", 0);
- cell->set_property("has_entry", false);
-#endif
- cell->signal_edited().connect(sigc::bind(sigc::mem_fun(*this, &Dialog_Import_CSV::on_field_edited), i));
- m_sample_view->append_column(*col);
- }
-
+ setup_sample_model(row);
Gtk::TreeModel::iterator iter = m_sample_model->append();
// -1 means the row to select target fields (see special handling in cell data funcs)
(*iter)[m_sample_columns.m_col_row] = -1;
@@ -945,12 +707,62 @@ void Dialog_Import_CSV::handle_line(const Glib::ustring& line, guint line_number
{
if(sample_rows < static_cast<guint>(m_sample_rows->get_value_as_int()))
{
- Gtk::TreeModel::iterator iter = m_sample_model->append();
- (*iter)[m_sample_columns.m_col_row] = m_rows.size() - 1;
+ Gtk::TreeModel::iterator tree_iter = m_sample_model->append();
+ (*tree_iter)[m_sample_columns.m_col_row] = m_parser->m_rows.size() - 1;
}
}
}
+void Dialog_Import_CSV::setup_sample_model(CsvParser::type_row_strings& row)
+{
+ m_sample_model = Gtk::ListStore::create(m_sample_columns);
+ m_sample_view->set_model(m_sample_model);
+
+ // Create field vector that contains the fields into which to import
+ // the data.
+ //m_fields.resize(row.size());
+ m_fields.resize(row.size());
+
+ // Start with a column showing the line number.
+ Gtk::CellRendererText* text = Gtk::manage(new Gtk::CellRendererText);
+ Gtk::TreeViewColumn* col = Gtk::manage(new Gtk::TreeViewColumn(_("Line")));
+ col->pack_start(*text, false);
+ col->set_cell_data_func(*text, sigc::mem_fun(*this, &Dialog_Import_CSV::line_data_func));
+ m_sample_view->append_column(*col);
+
+ for(guint i = 0; i < row.size(); ++ i)
+ {
+ m_sample_view->append_column(*Gtk::manage(column_factory(row[i], i)));
+ }
+}
+
+Gtk::TreeViewColumn* Dialog_Import_CSV::column_factory(const Glib::ustring& title, guint index)
+{
+ Gtk::TreeViewColumn* col = new Gtk::TreeViewColumn(title);
+ Gtk::CellRendererCombo* cell = cell_factory(index);
+ col->pack_start(*Gtk::manage(cell), true);
+ col->set_cell_data_func(*cell, sigc::bind(sigc::mem_fun(*this, &Dialog_Import_CSV::field_data_func), index));
+ col->set_sizing(Gtk::TREE_VIEW_COLUMN_AUTOSIZE);
+ return col;
+}
+
+Gtk::CellRendererCombo* Dialog_Import_CSV::cell_factory(guint index)
+{
+ Gtk::CellRendererCombo* cell = new Gtk::CellRendererCombo;
+#ifdef GLIBMM_PROPERTIES_ENABLED
+ cell->property_model() = m_field_model_sorted;
+ cell->property_text_column() = 0;
+ cell->property_has_entry() = false;
+#else
+ cell->set_property("model", m_field_model_sorted);
+ cell->set_property("text-column", 0);
+ cell->set_property("has_entry", false);
+#endif
+ cell->signal_edited().connect(sigc::bind(sigc::mem_fun(*this, &Dialog_Import_CSV::on_field_edited), index));
+
+ return cell;
+}
+
void Dialog_Import_CSV::line_data_func(Gtk::CellRenderer* renderer, const Gtk::TreeModel::iterator& iter)
{
const int row = (*iter)[m_sample_columns.m_col_row];
@@ -958,7 +770,7 @@ void Dialog_Import_CSV::line_data_func(Gtk::CellRenderer* renderer, const Gtk::T
#ifdef GLIBMM_EXCEPTIONS_ENABLED
if(!renderer_text)
throw std::logic_error("CellRenderer is not a CellRendererText in line_data_func");
-#endif
+#endif
if(row == -1)
renderer_text->set_property("text", Glib::ustring(_("Target Field")));
@@ -972,11 +784,11 @@ void Dialog_Import_CSV::field_data_func(Gtk::CellRenderer* renderer, const Gtk::
Gtk::CellRendererCombo* renderer_combo = dynamic_cast<Gtk::CellRendererCombo*>(renderer);
#ifdef GLIBMM_EXCEPTIONS_ENABLED
if(!renderer_combo) throw std::logic_error("CellRenderer is not a CellRendererCombo in field_data_func");
-#endif
+#endif
Glib::ustring text;
bool editable = false;
-
+
if(row == -1)
{
sharedptr<Field> field = m_fields[column_number];
@@ -991,17 +803,17 @@ void Dialog_Import_CSV::field_data_func(Gtk::CellRenderer* renderer, const Gtk::
{
// Convert to currently chosen field, if any, and back, too see how it
// looks like when imported:
-
+
if(column_number < m_fields.size())
{
sharedptr<Field> field = m_fields[column_number];
-
- if(row != -1 && (unsigned int)row < m_rows.size())
+
+ if(row != -1 && (unsigned int)row < m_parser->m_rows.size())
{
- const type_row_strings& row_strings = m_rows[row];
+ const CsvParser::type_row_strings& row_strings = m_parser->m_rows[row];
if(column_number < row_strings.size())
{
-
+
const Glib::ustring& orig_text = row_strings[column_number];
if(field)
@@ -1048,6 +860,10 @@ void Dialog_Import_CSV::field_data_func(Gtk::CellRenderer* renderer, const Gtk::
renderer_combo->set_property("editable", editable);
}
+
+/** Parse a row from a .cvs file. Note that this "line" might have newline
+ * characters inside one field value, inside quotes.
+ **/
void Dialog_Import_CSV::on_field_edited(const Glib::ustring& path, const Glib::ustring& new_text, guint column_number)
{
Gtk::TreeModel::Path treepath(path);
@@ -1086,18 +902,20 @@ void Dialog_Import_CSV::on_field_edited(const Glib::ustring& path, const Glib::u
}
}
-void Dialog_Import_CSV::set_state(State state)
+void Dialog_Import_CSV::set_parser_state(CsvParser::State state)
{
- if(m_state != state)
+ // Calling the member of a member, introduced by refactoring. TODO: clean up set_parser_state() interface.
+ if(m_parser->m_state != state)
{
- m_state = state;
+ m_parser->m_state = state;
+ // Should be emitted by parser?
m_signal_state_changed.emit();
}
}
void Dialog_Import_CSV::validate_primary_key()
{
- if(m_state == NONE || m_state == ENCODING_ERROR)
+ if(get_parser_state() == (CsvParser::NONE | CsvParser::ENCODING_ERROR))
{
m_error_label->hide();
set_response_sensitive(Gtk::RESPONSE_ACCEPT, false);
@@ -1115,7 +933,7 @@ void Dialog_Import_CSV::validate_primary_key()
// and the m_fields array is not up to date. It is set in handle_line()
// when the first line is parsed.
primary_key_selected = false;
- if(!m_rows.empty())
+ if(!m_parser->m_rows.empty())
{
for(type_vec_fields::iterator iter = m_fields.begin(); iter != m_fields.end(); ++ iter)
{
diff --git a/glom/dialog_import_csv.h b/glom/dialog_import_csv.h
index ca06b21..3b6e3f2 100644
--- a/glom/dialog_import_csv.h
+++ b/glom/dialog_import_csv.h
@@ -21,12 +21,12 @@
#ifndef GLOM_DIALOG_IMPORT_CSV_H
#define GLOM_DIALOG_IMPORT_CSV_H
+#include "import_csv.h"
#include <glom/base_db.h>
#include <memory>
#include <giomm/asyncresult.h>
#include <giomm/file.h>
-#include <giomm/inputstream.h>
#include <gtkmm/dialog.h>
#include <gtkmm/treeview.h>
#include <gtkmm/liststore.h>
@@ -46,20 +46,14 @@ class Dialog_Import_CSV
public:
typedef sigc::signal<void> SignalStateChanged;
- enum State {
- NONE,
- PARSING,
- ENCODING_ERROR,
- PARSED
- };
-
Dialog_Import_CSV(BaseObjectType* cobject, const Glib::RefPtr<Gtk::Builder>& builder);
void import(const Glib::ustring& uri, const Glib::ustring& into_table);
-
- State get_state() const { return m_state; }
- Glib::ustring get_target_table_name() const { return m_target_table->get_text(); }
- const Glib::ustring& get_filename() const { return m_filename; }
+
+ //TODO: move into parser?
+ CsvParser::State get_parser_state() const;
+ Glib::ustring get_target_table_name() const;
+ const Glib::ustring& get_filename() const;
unsigned int get_row_count() const;
unsigned int get_column_count() const;
@@ -85,20 +79,20 @@ private:
Glib::ustring get_current_encoding() const;
void begin_parse();
- void encoding_error();
+ void on_encoding_error();
bool on_idle_parse();
-
- /** Parse a row from a .cvs file. Note that this "line" might have newline
- * characters inside one field value, inside quotes.
- */
- void handle_line(const Glib::ustring& line, unsigned int line_number);
+
+ void on_line_scanned(const Glib::ustring& line, unsigned int line_number);
+ void setup_sample_model(CsvParser::type_row_strings& row);
+ Gtk::TreeViewColumn* column_factory(const Glib::ustring& title, guint index);
+ Gtk::CellRendererCombo* cell_factory(guint index);
void line_data_func(Gtk::CellRenderer* renderer, const Gtk::TreeModel::iterator& iter);
void field_data_func(Gtk::CellRenderer* renderer, const Gtk::TreeModel::iterator& iter, unsigned int column_number);
void on_field_edited(const Glib::ustring& path, const Glib::ustring& new_text, unsigned int column_number);
- void set_state(State state);
+ void set_parser_state(CsvParser::State state);
void validate_primary_key();
class EncodingColumns: public Gtk::TreeModelColumnRecord
@@ -127,6 +121,8 @@ private:
Gtk::TreeModelColumn<int> m_col_row;
};
+ std::auto_ptr<CsvParser> m_parser;
+
EncodingColumns m_encoding_columns;
Glib::RefPtr<Gtk::ListStore> m_encoding_model;
@@ -146,14 +142,8 @@ private:
Gtk::Label* m_error_label;
Glib::RefPtr<Gio::File> m_file;
- Glib::RefPtr<Gio::FileInputStream> m_stream;
Glib::ustring m_filename;
- // The raw data in the original encoding. We keep this so we can convert
- // from the user-selected encoding to UTF-8 every time the user changes
- // the encoding.
- std::vector<char> m_raw;
-
struct Buffer {
char buf[1024];
};
@@ -164,29 +154,6 @@ private:
// auto-detection is disabled.
int m_auto_detect_encoding;
- // We use the low-level Glib::IConv routines to progressively convert the
- // input data in an idle handler.
- class Parser
- {
- public:
- Parser(const char* encoding): conv("UTF-8", encoding), input_position(0), line_number(0) {}
- ~Parser() { idle_connection.disconnect(); }
-
- Glib::IConv conv;
- std::vector<char>::size_type input_position;
- std::string current_line;
- sigc::connection idle_connection;
- unsigned int line_number;
- };
-
- std::auto_ptr<Parser> m_parser;
- State m_state;
-
- // Parsed data:
- typedef std::vector<Glib::ustring> type_row_strings;
- typedef std::vector<type_row_strings> type_rows;
- type_rows m_rows;
-
// The fields into which to import the data:
typedef std::vector< sharedptr<Field> > type_vec_fields;
type_vec_fields m_fields;
diff --git a/glom/dialog_import_csv_progress.cc b/glom/dialog_import_csv_progress.cc
index ed26c34..9296fa0 100644
--- a/glom/dialog_import_csv_progress.cc
+++ b/glom/dialog_import_csv_progress.cc
@@ -57,16 +57,16 @@ void Dialog_Import_CSV_Progress::import(Dialog_Import_CSV& data_source)
m_data_source = &data_source;
m_current_row = 0;
- switch(data_source.get_state())
+ switch(data_source.get_parser_state())
{
- case Dialog_Import_CSV::PARSING:
+ case CsvParser::PARSING:
// Wait for the parsing to finish. We do not start importing before the file has been
// parsed completely since we would not to rollback our changes in case of a
// parsing error.
m_progress_bar->set_text(Glib::ustring::compose(_("Parsing CSV file %1"), data_source.get_filename()));
m_ready_connection = data_source.signal_state_changed().connect(sigc::mem_fun(*this, &Dialog_Import_CSV_Progress::on_state_changed));
break;
- case Dialog_Import_CSV::PARSED:
+ case CsvParser::PARSED:
begin_import();
break;
default:
@@ -108,11 +108,11 @@ void Dialog_Import_CSV_Progress::on_state_changed()
{
switch(m_data_source->get_state())
{
- case Dialog_Import_CSV::ENCODING_ERROR:
+ case CsvParser::ENCODING_ERROR:
// Cancel on error
response(Gtk::RESPONSE_CANCEL);
break;
- case Dialog_Import_CSV::PARSED:
+ case CsvParser::PARSED:
// Begin importing when fully parsed
begin_import();
//m_progress_connection = Glib::signal_idle().connect(sigc::mem_fun(*this, &Dialog_Import_CSV_Progress::on_idle_import));
diff --git a/glom/import_csv.cc b/glom/import_csv.cc
new file mode 100644
index 0000000..8844500
--- /dev/null
+++ b/glom/import_csv.cc
@@ -0,0 +1,314 @@
+/* Glom
+ *
+ * Copyright (C) 2001-2004 Murray Cumming
+ * Copyright (C) 2009 Openismus GmbH
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 02111-1307, USA.
+ */
+
+#include "import_csv.h"
+
+#include <cerrno>
+
+// On Windows, "iconv" seems to be a define for "libiconv", breaking the Glib::IConv::iconv() call.
+#ifdef iconv
+#undef iconv
+#endif
+
+namespace Glom
+{
+
+bool CsvParser::next_char_is_quote(const Glib::ustring::const_iterator& iter, const Glib::ustring::const_iterator& end)
+{
+ if(iter == end)
+ return false;
+
+ // Look at the next character to see if it's really "" (an escaped "):
+ Glib::ustring::const_iterator iter_next = iter;
+ ++iter_next;
+ if(iter_next != end)
+ {
+ const gunichar c_next = *iter_next;
+ if(c_next == CsvParser::QUOTE)
+ {
+ return true;
+ }
+ }
+
+ return false;
+}
+
+CsvParser::CsvParser(const char* encoding)
+: m_raw(0),
+ m_encoding(encoding),
+ m_input_position(0),
+ m_idle_connection(),
+ m_line_number(0),
+ m_state(NONE),
+ m_stream(),
+ m_rows()
+{}
+
+CsvParser::~CsvParser()
+{
+ m_idle_connection.disconnect();
+}
+
+CsvParser::SignalEncodingError CsvParser::signal_encoding_error() const
+{
+ return m_encoding_error;
+}
+
+CsvParser::SignalLineScanned CsvParser::signal_line_scanned() const
+{
+ return m_line_scanned;
+}
+
+void CsvParser::set_encoding(const char* encoding)
+{
+ m_encoding = static_cast<std::string>(encoding);
+}
+
+// Parse the field in a comma-separated line, returning the field including the quotes:
+// (But can it operate on non-UTF, read: binary, data?)
+Glib::ustring::const_iterator CsvParser::advance_field(const Glib::ustring::const_iterator& iter, const Glib::ustring::const_iterator& end, Glib::ustring& field)
+{
+ bool inside_quotes = false;
+ //bool string_finished = false; //Ignore anything after "something", such as "something"else,
+
+ field.clear();
+
+ Glib::ustring::const_iterator walk;
+ for(walk = iter; walk != end; ++walk)
+ {
+ const gunichar c = *walk;
+
+ //if(string_finished)
+ // continue;
+
+ if(inside_quotes)
+ {
+ // End of quoted string?
+ if(c == CsvParser::QUOTE)
+ {
+ if(CsvParser::next_char_is_quote(walk, end))
+ {
+ // This is "" so it's not an end quote. Just add one quote:
+ field += c;
+ ++walk; //Skip the second ".
+ }
+ else
+ {
+ inside_quotes = false;
+ //string_finished = true; //Ignore anything else before the next comma.
+ }
+
+ continue;
+ }
+ }
+ else
+ {
+ // Start of quoted string:
+ if((c == CsvParser::QUOTE))
+ {
+ inside_quotes = true;
+ continue;
+ }
+ // End of field:
+ else if(!inside_quotes && c == CsvParser::DELIMITER)
+ {
+ break;
+ }
+
+ continue;
+ }
+
+ field += c; // Just so that we don't need to iterate through the field again, since there is no Glib::ustring::substr(iter, iter)
+ }
+
+ // TODO: Throw error if still inside a quoted string?
+ std::cout << "debug: field=" << field << std::endl;
+ return walk;
+}
+
+void CsvParser::clear()
+{
+ //m_stream.reset();
+ //m_raw.clear();
+ m_rows.clear();
+ // Set to current encoding I guess ...
+ //m_conv("UTF-8", encoding),
+ m_input_position= 0;
+ // Disconnect signal handlers, too? Nah, I don't think so ...
+ //m_idle_connection.disconnect();
+ m_line_number = 0;
+ m_state = NONE;
+}
+
+bool CsvParser::on_idle_parse()
+{
+ Glib::IConv conv("UTF-8", m_encoding);
+
+ // The amount of bytes to process in one pass of the idle handler:
+ static const guint CONVERT_BUFFER_SIZE = 1024;
+
+ const char* inbuffer = &m_raw[m_input_position];
+ char* inbuf = const_cast<char*>(inbuffer);
+ gsize inbytes = m_raw.size() - m_input_position;
+ char outbuffer[CONVERT_BUFFER_SIZE];
+ char* outbuf = outbuffer;
+ gsize outbytes = CONVERT_BUFFER_SIZE;
+
+ const std::size_t result = conv.iconv(&inbuf, &inbytes, &outbuf, &outbytes);
+ bool more_to_process = (inbytes != 0);
+
+ if(result == static_cast<size_t>(-1))
+ {
+ if(errno == EILSEQ)
+ {
+ // Invalid text in the current encoding.
+ signal_encoding_error().emit();
+ return false;
+ }
+
+ // If EINVAL is set, this means that an incomplete multibyte sequence was at
+ // the end of the input. We might have some more bytes, but those do not make
+ // up a whole character, so we need to wait for more input.
+ if(errno == EINVAL)
+ {
+ if(!m_stream)
+ {
+ // This means that we already reached the end of the file. The file
+ // should not end with an incomplete multibyte sequence.
+ signal_encoding_error().emit();
+ return false;
+ }
+ else
+ {
+ more_to_process = false;
+ }
+ }
+ }
+
+ m_input_position += (inbuf - inbuffer);
+
+ // We now have outbuf - outbuffer bytes of valid UTF-8 in outbuffer.
+ const char* prev_line_end = outbuffer;
+ const char* prev = prev_line_end;
+
+ // Identify the record rows in the .csv file.
+ // We can't just search for newlines because they may be inside quotes too.
+ // TODO: Use a regex instead, to more easily handle quotes?
+ bool in_quotes = false;
+ while(true)
+ {
+ // Note that, unlike std::string::find*, std::find* returns an iterator (char*), not a position.
+ // It returns outbuf if none is found.
+ const char newline_to_find[] = { '\r', '\n', '\0' };
+ const char* pos_newline = std::find_first_of<const char*>(prev, outbuf, newline_to_find, newline_to_find + sizeof(newline_to_find));
+
+ const char quote_to_find[] = {(char)QUOTE};
+ const char* pos_quote = std::find_first_of<const char*>(prev, outbuf, quote_to_find, quote_to_find + sizeof(quote_to_find));
+
+ // Examine the first character (quote or newline) that was found:
+ const char* pos = pos_newline;
+ if((pos_quote != outbuf) && pos_quote < pos)
+ pos = pos_quote;
+
+ if(pos == outbuf)
+ break;
+
+ char ch = *pos;
+
+ if(ch == '\0')
+ {
+ // There is a null byte in the conversion. Because normal text files don't
+ // contain null bytes this only occurs when converting, for example, a UTF-16
+ // file from ISO-8859-1 to UTF-8 (note that the UTF-16 file is valid ISO-8859-1 -
+ // it just contains lots of nullbytes). We therefore produce an error here.
+ signal_encoding_error().emit();
+ return false;
+ }
+ else if(in_quotes)
+ {
+ // Ignore newlines inside quotes.
+
+ // End quote:
+ if(ch == (char)QUOTE)
+ in_quotes = false;
+
+ prev = pos + 1;
+ continue;
+ }
+ else
+ {
+ // Start quote:
+ if(ch == (char)QUOTE)
+ {
+ in_quotes = true;
+ prev = pos + 1;
+ continue;
+ }
+
+ // Found a newline (outside of quotes) that marks the end of the line:
+ m_current_line.append(prev_line_end, pos - prev_line_end);
+ ++(m_line_number);
+
+ if(!m_current_line.empty())
+ {
+ // Emitting a signal frees the parser of having to know the line
+ // handling object instance (inversion of control using the observer
+ // pattern).
+ signal_line_scanned().emit(m_current_line, m_line_number);
+ }
+
+ m_current_line.clear();
+
+ // Skip linebreak
+ prev = pos + 1;
+
+ // Skip DOS-style linebreak (\r\n)
+ if(ch == '\r'
+ && prev != outbuf && *prev == '\n')
+ {
+ ++prev;
+ }
+
+ prev_line_end = prev;
+ }
+ }
+
+ // Append last chunk of this line
+ m_current_line.append(prev, outbuf - prev);
+ if(!m_stream && m_raw.size() == m_input_position)
+ {
+ ++(m_line_number);
+
+ // Handle last line, if nonempty
+ if(m_current_line.empty())
+ {
+ signal_line_scanned().emit(m_current_line, m_line_number);
+ }
+
+ // Parsed whole file, done
+ m_state = PARSED;
+ }
+
+ // Continue if there are more bytes to process
+ return more_to_process;
+}
+
+} // namespace Glom
diff --git a/glom/import_csv.h b/glom/import_csv.h
new file mode 100644
index 0000000..380f12d
--- /dev/null
+++ b/glom/import_csv.h
@@ -0,0 +1,104 @@
+/* Glom
+ *
+ * Copyright (C) 2001-2004 Murray Cumming
+ * Copyright (C) 2009 Openismus GmbH
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 02111-1307, USA.
+ */
+
+#ifndef GLOM_IMPORT_CSV_H
+#define GLOM_IMPORT_CSV_H
+
+#include "base_db.h"
+
+#include <memory>
+#include <giomm/asyncresult.h>
+#include <giomm/file.h>
+#include <giomm/inputstream.h>
+#include <gtkmm/liststore.h>
+
+namespace Glom
+{
+
+// We use the low-level Glib::IConv routines to progressively convert the
+// input data in an idle handler.
+
+class CsvParser
+{
+public:
+ explicit CsvParser(const char* encoding);
+
+ ~CsvParser();
+
+ // TODO: prefix enums with STATE_* ?
+ enum State {
+ NONE,
+ PARSING,
+ ENCODING_ERROR,
+ PARSED
+ };
+
+ static const gunichar DELIMITER = ',';
+ static const gunichar QUOTE = '\"';
+
+ static bool next_char_is_quote(const Glib::ustring::const_iterator& iter, const Glib::ustring::const_iterator& end);
+ static Glib::ustring::const_iterator advance_field(const Glib::ustring::const_iterator& iter, const Glib::ustring::const_iterator& end, Glib::ustring& field);
+
+ // Signals
+ typedef sigc::signal<void> SignalEncodingError;
+ SignalEncodingError signal_encoding_error() const;
+
+ typedef sigc::signal<void, std::string, unsigned int> SignalLineScanned;
+ SignalLineScanned signal_line_scanned() const;
+
+ // Make parser object reusable.
+ void clear();
+
+ // In order to not make the UI feel sluggish during larger imports we parse
+ // on chunk at a time in the idle handler.
+ bool on_idle_parse();
+
+ void set_encoding(const char* encoding);
+
+//For now, everything's public. Until we know how our interface will look like.
+//private:
+ // The raw data in the original encoding. We keep this so we can convert
+ // from the user-selected encoding to UTF-8 every time the user changes
+ // the encoding.
+ std::vector<char> m_raw;
+
+ std::string m_encoding;
+ std::vector<char>::size_type m_input_position;
+ std::string m_current_line;
+ sigc::connection m_idle_connection;
+ unsigned int m_line_number;
+ State m_state;
+ Glib::RefPtr<Gio::FileInputStream> m_stream;
+
+ // Parsed data:
+ typedef std::vector<Glib::ustring> type_row_strings;
+ typedef std::vector<type_row_strings> type_rows;
+ type_rows m_rows;
+
+ SignalEncodingError m_encoding_error;
+ SignalLineScanned m_line_scanned;
+};
+
+} //namespace Glom
+
+#endif //GLOM_IMPORT_CSV_H
+
+
[
Date Prev][
Date Next] [
Thread Prev][
Thread Next]
[
Thread Index]
[
Date Index]
[
Author Index]