Remove the apparently bogus "FALSE &&" condition in SC_forget_unnamed(). #1
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…ore than by VC+7.
Per report Stephen M. Lathrop.
Per report Stephen M. Lathrop.
2. Support column alias without "as" so that links from the SQLServer work. 3. Take ';' into account when the driver adds "for read only" clause. 4. Use the E'.. ' notation not only in '=' expressions but also in LIKE expressions. 5. Change to return milliseconds parts for timestamp fields. 6. Change to return a specific sqlstate in case of multiple parameters.
(PostgreSQL Version 8.3 of MSVC)
8.2 -> 8.3
1. SQLGUID type support thanks to Jan-Willem Goossens. 2. Fix a bug about silently adding a *for read only* clause. 3. Fix a 64bit mode bug about handling of arrays of parameters. 4. Change the implemetatin of SQLForeignKeys() for 8.3+ servers. 5. Not commit the transaction too early in useDeclareFetch mode. 6. Add a cursor open check for SQLPrepare().
…folder on some versions of Windows
1. Reset the column binding information after SQLMoreResults(). 2. Save the rowset size properly for the FETCH_NEXT operation in case of >= 3.0 drivers. Enable geqo optiomizer by default.
2. Check strerror_r function's return type. 3. Suppress some compiler warnings.
We liberally assign between SQLCHAR * and char *, which generates a lot of compiler warnings about different pointer signedness. Seems like a lost cause to add casts everywhere (and it might not be good for robustness anyway, because it might mask a bug where the datatype of one side is changed to a struct or something in the future), so just silence the warnings.
Compiler warned about this.
Silences a compiler warning.
For some reason, the destructor was not being called when compiled with gcc. Not sure why, but I'm assuming it was an oversight. This silences a compiler warning complaining that finalize_global_cs() function was unused.
BTW, the error and notice handling could be made much smarter wrt. truncation. The caller of handle_notice/error_message knows the message size in advance, so it could easily pass it down and we could allocate a buffer of the right size, instead of using a fixed size buffer that can be too small.
If realloc fails and returns NULL, don't forget the old array. Also check for overflow of num_Stmts, which is only a 16-bit int
Found by Coverity scan.
We reference the same pointer without null-check elsewhere in the function, so if it's ever NULL, we'll crash anyway. (but looking at the callers, it never is NULL) Found by Coverity scan.
The problem is that if the loop above doesn't find a result set with command tag "FETCH ...", res is left to point to an already-freed result set, and we'll go ahead and call QR_set_withhold on the already-freed struct. That can't happen in practice, because the backend does always return a result set with "FETCH ..." command tag when a FETCH statement is issued, but better safe than sorry.
Found by Coverity scan.
Coverity doesn't like the fact that the struct is passed to SC_set_rowset_start without initializing rowset_start field. It's OK, SC_set_rowset_start doesn't actually use the 'incr' value calculated from the field if the 3rd arg is FALSE, so this is a false positive, but seems like a good idea to be tidy here anyway.
This avoids accessing unitialized bitmap[0]. It was harmless, as the uninitialize value that was read was not used for anything, but Coverity complained about it.
It's a good habit to mark constants as such. But I'm also hoping that this would silence Coverity from complaining that ver2str and ver3str might not fit in 6 bytes.
It wasn't actually true, but apparently not all compiler versions were smart enough to realize it.
To silence a Coverity warning that the semicolon might be misplaced.
If you have a lot of chained results, e.g. if you array bind a lot of parameters, recursing through the whole chain might make you run out of stack space. Convert recursion to iteration. Also, when we're freeing all the results in the chain, we only need to call QR_set_cursor(self, NULL) once, on the first result in the chain. QR_set_cursor itself loops through the whole chain, resetting the cursor name of every result. By only calling it for the first result, we avoid O(n^2) behavior, significantly speeding up SQLFreeHandle() on a statement with a large number of chained results.
There was at least one real bug caused by the multiple evaluation of 2nd argument to SC_set_result(). SC_pre_execute() passed QR_Constructor() call as the 2nd argument, so we ended up creating extra result sets that were leaked.
…depaths. This patch eliminates O(n^2) behavior on the number of chained results in SC_execute. This makes it feasible to deal with large parameter arrays, for example. The idea is to track the owner statement of each result object, so that when we need to check if a result object is already in the chain of results of a statement, we just need to look at the owner field, not loop through the whole chain. Also, keep track of the last result in the chain in StatementClass, so that we don't need to iterate through it when appending to the end.
If a query was sent to the server using the Parse command, we returned the dummy QResult for the Parse, with 0 rows, to the caller of SQLExecute. Fix by clearing out the result set just before executing.
…wrong. With the bogus fix, we only returned the last result set for an array bound statement. Move resetting the result set to before the next_param_row label, so that it's only done once when using a parameter array.
This was added, perhaps accidentally, in 5f35a04, and resulted in no results being returned from SQLFetch() for prepared statements.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This was added, perhaps accidentally, in 5f35a04, and resulted in no results
being returned from SQLFetch() for prepared statements.