Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors...

16
Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects Authors: Evgeniy Ryzhkov Date: 20.04.2011 Abstract The article demonstrates errors detected with the static code analyzer integrated into Visual Studio 2010. The research was performed on five open source projects. The same projects were also checked with PVS-Studio. Results of comparing these two tools are presented at the end of the article. Introduction The article "Difficulties of comparing code analyzers, or don't forget about usability" [1 ] tells that it is not so easy to compare two tools as it may seem because the parameter of usability is also highly significant besides the technical characteristics proper. Still we cannot do without comparing tools by errors they can detect. Of course, there is no sense in just calculating the number of errors. So we decided to carry out a practical experiment of error detection in real projects. We checked five random open source projects with the static analyzer integrated into Visual Studio 2010 Premium. We looked through the whole message list and chose explicit errors. Then we made the same steps with PVS-Studio. Here is a list of projects which participated in the research: eMule Plus ; Pixie ; VirtualDub ; WinMerge ; XUIFramework . Let's go! eMule Plus The total number of messages generated by the Visual Studio static analyzer is 237, 4 of them being real errors. The total number of messages generated by PVS-Studio is 68, 3 of them being real errors.

description

The article demonstrates errors detected with the static code analyzer integrated into Visual Studio 2010. The research was performed on five open source projects. The same projects were also checked with PVS-Studio. Results of comparing these two tools are presented at the end of the article.

Transcript of Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors...

Page 1: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

Comparing the general static analysis in

Visual Studio 2010 and PVS-Studio by

examples of errors detected in five open

source projects

Authors: Evgeniy Ryzhkov

Date: 20.04.2011

Abstract The article demonstrates errors detected with the static code analyzer integrated into Visual Studio

2010. The research was performed on five open source projects. The same projects were also checked

with PVS-Studio. Results of comparing these two tools are presented at the end of the article.

Introduction The article "Difficulties of comparing code analyzers, or don't forget about usability" [1] tells that it is

not so easy to compare two tools as it may seem because the parameter of usability is also highly

significant besides the technical characteristics proper.

Still we cannot do without comparing tools by errors they can detect. Of course, there is no sense in just

calculating the number of errors. So we decided to carry out a practical experiment of error detection in

real projects.

We checked five random open source projects with the static analyzer integrated into Visual Studio

2010 Premium. We looked through the whole message list and chose explicit errors. Then we made the

same steps with PVS-Studio.

Here is a list of projects which participated in the research:

• eMule Plus;

• Pixie;

• VirtualDub;

• WinMerge;

• XUIFramework.

Let's go!

eMule Plus The total number of messages generated by the Visual Studio static analyzer is 237, 4 of them being real

errors.

The total number of messages generated by PVS-Studio is 68, 3 of them being real errors.

Page 2: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

Visual Studio: warning C6054: String 'szwThemeFile' might not be

zero-terminated. c:\emuleplus\dialogmintraybtn.hpp 445

WCHAR szwThemeFile[MAX_PATH];

WCHAR szwThemeColor[256];

if (m_themeHelper.GetCurrentThemeName(szwThemeFile,

ARRSIZE(szwThemeFile), szwThemeColor,

ARRSIZE(szwThemeColor), NULL, 0) != S_OK)

return NULL;

WCHAR *p;

if ((p = wcsrchr(szwThemeFile, L'\\')) == NULL)

Indeed, a line may not end with 0, which will cause potential problems. But in this particular case, this

error is not likely to reveal itself.

Visual Studio: warning C6269: Possibly incorrect or der of operations:

dereference ignored. c:\emuleplus\customautocomplet e.cpp 277

PVS-Studio: V532 Consider inspecting the statement of '*pointer++'

pattern. Probably meant: '(*pointer)++'. customauto complete.cpp 277

if (pceltFetched != NULL)

*pceltFetched++;

The programmer here "is not good at" using the (*ptr)++ expression. Although this construct seems to

be rather safe, still this error is widespread.

Visual Studio: warning C6298: Argument '6': using a read-only string

as a writable string argument. This will attempt to write into static

read-only memory and cause random crashes.

c:\emuleplus\firewallopener.cpp 183

Page 3: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

HRESULT hr = pNSC->AddPortMapping(

riPortRule.m_strRuleName.AllocSysString(), riPort Rule.m_byProtocol,

riPortRule.m_nPortNumber, riPortRule.m_nPortNumbe r, 0, L"127.0.0.1",

ICSTT_IPADDRESS, &pNSPM);

Although it is not an error, the analyzer-generated message is fair. In general, this is a problem of all the

static analyzers. They produce absolutely correct messages but they are far not always real errors. Does

it mean that such tools and messages are useless? No, it does not, because fixing even such warnings

helps to increase an overall quality of code.

Visual Studio: warning C6314: Incorrect order of op erations: bitwise-

or has higher precedence than the conditional-expre ssion operator.

Add parentheses to clarify intent. c:\emuleplus\sea rchlistctrl.cpp 659

PVS-Studio: V502 Perhaps the '?:' operator works in a different way

than it was expected. The '?:' operator has a lower priority than the

'|' operator. searchlistctrl.cpp 659

menuSearchFile.AppendMenu( MF_STRING |

((iSelectionMark != -1) && (dwSelectedCount > 0) &&

g_App.m_pServerConnect->IsConnected() &&

((pCurServer = g_App.m_pServerConnect->GetCurrent Server())!= NULL)&&

(pCurServer->GetTCPFlags() & SRV_TCPFLG_RELATEDSE ARCH)) ?

MF_ENABLED : MF_GRAYED, MP_SEARCHRELATED,

GetResString(IDS_SEARCHRELATED));

Here (because of the construct's complexity) we have wrong priorities of operators. For how long it has

been said... Who prevented the programmer from arranging this code in several separate expressions

instead of writing everything in one line (as it was done in the original program)? No, programmers

would always want to "write it smart".

PVS-Studio: V519 The 'm_clrSample' object is assign ed values twice

successively. Perhaps this is a mistake. fontprevie wcombo.cpp 61

Page 4: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

CFontPreviewCombo::CFontPreviewCombo()

{

...

m_clrSample = GetSysColor(COLOR_WINDOWTEXT);

m_clrSample = RGB(60,0,0);

...

}

Perhaps the developers wanted to see how the RGB(60,0,0) color would look and forgot to remove it.

Pixie The total number of messages generated by the Visual Studio static analyzer is 18, 0 of them being real

errors.

The total number of messages generated by PVS-Studio is 65, 5 of them being real errors.

PVS-Studio: V519 The 'numRays' object is assigned v alues twice

successively. Perhaps this is a mistake. bundles.cp p 579

void CGatherBundle::post() {

numRays = last;

numRays = 0;

last = 0;

depth++;

}

Oh, how suspicious it is when numRays is first initialized by one value and then by another. Is it an error

or not? Only the code's author knows exactly. But it alerts me!

PVS-Studio: V501 There are identical sub-expression s to the left and

to the right of the '|' operator: PARAMETER_DPDU | PARAMETER_DPDU

quadrics.cpp 880

if (up & (PARAMETER_DPDU | PARAMETER_DPDU)) {

Page 5: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

There must be something else here. By the way, here you a general note on fixing errors detected by a

static analyzer. While in some cases correction is obvious and anyone can fix an error, in some other

cases only the author of the code can make out what exactly was intended there. It is on the question if

one could create a tool that can correct errors "like in Word".

PVS-Studio: V501 There are identical sub-expression s to the left and

to the right of the '|' operator: SLC_VECTOR | SLC_ VECTOR

expression.cpp 2604

lock(N, getConversion(SLC_VECTOR | SLC_VECTOR,param eters[2]));

SLC_VECTOR mentioned twice in such a context certainly signals an error.

PVS-Studio: V505 The 'alloca' function is used insi de the loop.

This can quickly overflow stack. polygons.cpp 1120

inline void triangulatePolygon(...) {

...

for (i=1;i<nloops;i++) {

...

do {

...

do {

...

CTriVertex *snVertex = (CTriVertex *)

alloca(2*sizeof(CTriVertex));

...

} while(dVertex != loops[0]);

...

} while(sVertex != loops[i]);

...

}

Page 6: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

...

}

Having deep nesting, the alloca call might cause stack overflow.

VirtualDub The total number of messages generated by the Visual Studio static analyzer is 24 messages, 0 of them

being real errors.

The total number of messages generated by PVS-Studio is 41, 2 of them being real errors.

PVS-Studio: V547 Expression 'c < 0' is always false .

Unsigned type value is never < 0. lexer.cpp 279

typedef unsigned short wint_t;

wint_t lexgetescape() {

wint_t c = lexgetc();

if (c < 0)

fatal("Newline found in escape sequence");

...

}

The code will never be called because the expression is always false.

PVS-Studio: V557 Array overrun is possible. The '9' index is pointing

beyond array bound. f_convolute.cpp 73

struct ConvoluteFilterData {

long m[9];

long bias;

void *dyna_func;

DWORD dyna_size;

DWORD dyna_old_protect;

Page 7: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

BOOL fClip;

};

static unsigned long __fastcall do_conv(

unsigned long *data,

const ConvoluteFilterData *cfd,

long sflags, long pit)

{

long rt0=cfd->m[9], gt0=cfd->m[9], bt0=cfd->m[9];

...

}

A trivial array overflow.

WinMerge The total number of messages generated by the Visual Studio static analyzer is 343, 3 of them being real

errors.

The total number of messages generated by PVS-Studio is 69, 12 of them being real errors.

Visual Studio: warning C6313: Incorrect operator: z ero-valued flag

cannot be tested with bitwise-and. Use an equality test to check for

zero-valued flags. c:\winmerge\src\bcmenu.cpp 1489

else if (nFlags&MF_STRING){

ASSERT(!(nFlags&MF_OWNERDRAW));

ModifyMenu(pos,nFlags,nID,mdata->GetString());

}

Not very lucky condition. Of course, the programmer wanted to write something different, but it

happened that way.

Visual Studio: warning C6287: Redundant code: the l eft and right

sub-expressions are identical.

c:\winmerge\src\editlib\ccrystaleditview.cpp 1560

Page 8: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

PVS-Studio: V501 There are identical sub-expression s to the left and

to the right of the '||' operator: c == '}' || c == '}'

ccrystaleditview.cpp 1560

bool

isclosebrace (TCHAR c)

{

return c == _T ('}') || c == _T ('}') || c == _T (']')

|| c == _T ('>');

}

Not all the parentheses types are checked. Why? It is usual that "Copy-paste-technology" leads to

errors.

Visual Studio: warning C6287: Redundant code: the l eft and right

sub-expressions are identical. c:\winmerge\src\merg edoc.cpp 1165

PVS-Studio: V501 There are identical sub-expression s to the left and

to the right of the '||' operator. mergedoc.cpp 116 5

if ((m_nBufferType[nBuffer] == BUFFER_UNNAMED) ||

(m_nBufferType[nBuffer] == BUFFER_UNNAMED))

nSaveErrorCode = SAVE_NO_FILENAME;

Another unlucky condition and again it seems to be the copy-paste's fault.

PVS-Studio: V551 The code under this 'case' label i s unreachable.

The value range of signed char type: [-128, 127].

ccrystaltextview.cpp 1646

TCHAR ch = strText[i];

Page 9: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

switch (ch)

{

case 0xB7:

case 0xBB:

strHTML += ch;

strHTML += _T("<wbr>");

bLastCharSpace = FALSE;

nNonbreakChars = 0;

break;

And here we have a sample of code which will never be used. Everything seems alright, case is written

and all, but it will never get control because the value range is too narrow. TCHAR in this case is the char

type.

PVS-Studio: V524 It is odd that the body of 'IsVali dTextPosX' function

is fully equivalent to the body of 'IsValidTextPos' function

(ccrystaltextview.cpp, line 3700). ccrystaltextview .cpp 3707

bool CCrystalTextView::IsValidTextPos (const CPoint &point)

{

return GetLineCount () > 0 && m_nTopLine >= 0 &&

m_nOffsetChar >= 0 && point.y >= 0 &&

point.y < GetLineCount () && point.x >= 0 &&

point.x <= GetLineLength (point.y);

}

bool CCrystalTextView::IsValidTextPosX (const CPoin t &point)

{

return GetLineCount () > 0 && m_nTopLine >= 0 &&

m_nOffsetChar >= 0 && point.y >= 0 &&

point.y < GetLineCount () && point.x >= 0 &&

point.x <= GetLineLength (point.y);

Page 10: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

}

bool CCrystalTextView::IsValidTextPosY (const CPoin t &point)

{

return GetLineCount () > 0 && m_nTopLine >= 0 &&

m_nOffsetChar >= 0 && point.y >= 0 &&

point.y < GetLineCount ();

}

These are very similar functions... The developers copy-pasted again and again and forgot to fix the

result. The IsValidTextPosX() function performs an excess check.

PVS-Studio: V563 It is possible that this 'else' br anch must apply to

the previous 'if' statement. bcmenu.cpp 1852

if(IsLunaMenuStyle())

if(!xp_space_accelerators)return;

else

if(!original_space_accelerators)return;

Who can look at the code and say exactly to what if else refers? Was it the thing the programmer

wanted to do? Are you sure?

PVS-Studio: V556 The values of different enum types are compared:

switch(ENUM_TYPE_A) { case ENUM_TYPE_B: ... }. diff wrapper.cpp 956

enum output_style {}

...

enum DiffOutputType m_outputStyle;

switch (m_options.m_outputStyle)

{

Page 11: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

case OUTPUT_CONTEXT:

enum and switch types were a bit mixed up. But it's alright, isn't it?

PVS-Studio: V530 The return value of function 'empt y' is required to

be utilized. diractions.cpp 1307

void CDirView::GetItemFileNames(int sel, String& st rLeft,

String& strRight) c onst

{

UINT_PTR diffpos = GetItemKey(sel);

if (diffpos == (UINT_PTR)SPECIAL_ITEM_POS)

{

strLeft.empty();

strRight.empty();

The case when empty() does not empty. This is an example of an extremely wrong name for a method.

PVS-Studio: V524 It is odd that the body of 'OnUpda teLastdiff'

function is fully equivalent to the body of 'OnUpda teFirstdiff'

function (DirView.cpp, line 2189). dirview.cpp 2220

void CDirView::OnUpdateLastdiff(CCmdUI* pCmdUI)

{

int firstDiff = GetFirstDifferentItem();

if (firstDiff > -1)

pCmdUI->Enable(TRUE);

else

pCmdUI->Enable(FALSE);

}

void CDirView::OnUpdateFirstdiff(CCmdUI* pCmdUI)

Page 12: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

{

int firstDiff = GetFirstDifferentItem();

if (firstDiff > -1)

pCmdUI->Enable(TRUE);

else

pCmdUI->Enable(FALSE);

}

Two more very similar functions...

PVS-Studio: V501 There are identical sub-expression s

'pView1->GetTextBufferEol (line)' to the left and t o the right of

the '!=' operator. mergedoclinediffs.cpp 216

if (pView1->GetTextBufferEol(line) !=

pView1->GetTextBufferEol(line))

{

Either this or that... Or not? Perhaps there must be pView2 here.

PVS-Studio: V530 The return value of function 'empt y' is required to

be utilized. varprop.cpp 133

void VariantValue::Clear()

{

m_vtype = VT_NULL;

m_bvalue = false;

m_ivalue = 0;

m_fvalue = 0;

m_svalue.empty();

m_tvalue = 0;

}

Page 13: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

Again empty() does not empty the string at all.

PVS-Studio: V510 The 'Format' function is not expe cted to receive

class-type variable as 'N' actual argument". PropSh el 105

String GetSysError(int nerr);

...

CString msg;

msg.Format(

_T("Failed to open registry key HKCU/%s:\n\t%d : %s"),

f_RegDir, retVal, GetSysError(retVal)

);

When various emergencies occur, WinMerge will try to inform the user about errors but in some cases it

will fail. At first sight everything looks OK but actually the "String" type is just "std::wstring". Therefore

we will print rubbish at best or get an Access Violation error at worst. The correct code must have a call

of c_str().

PVS-Studio: V534 It is likely that a wrong variabl e is being compared inside the 'for' operator. Consider reviewing 'i'." BinTrans.cpp 357

// Get length of translated array of bytes from tex t.

int Text2BinTranslator::iLengthOfTransToBin(

char* src, int srclen )

{

...

for (k=i; i<srclen; k++)

{

if (src[k]=='>')

break;

}

...

Page 14: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

}

The analyzer found a suspicious loop. This code is prone to Access Violation. The loop must go on until it

finds the '>' character or a string with the length of 'srclen' characters comes to an end. But the

programmer by accident used the 'k' variable instead of 'i' for comparison. If the '>' character is not

found, everything will be sad.

XUIFramework The total number of messages generated by the Visual Studio static analyzer is 93, 2 of them being real

errors.

The total number of messages generated by PVS-Studio is 30, 5 of them being real errors.

Visual Studio: warning C6269: Possibly incorrect or der of operations:

dereference ignored

c:\xui-gui framework\widgets\cstatichtml\ppdrawmana ger.cpp 298

PVS-Studio: V532 Consider inspecting the statement of '*pointer++'

pattern. Probably meant: '(*pointer)++'. ppdrawmana ger.cpp 298

for (DWORD pixel = 0; pixel < dwWidth * dwHeight; p ixel++, *pBits++)

Again the programmer is not good at using *ptr++. As I have said above, this is a widespread error.

Visual Studio: warning C6283: 'pBuffer' is allocate d with array new[],

but deleted with scalar delete.

c:\xui-gui framework\widgets\cxstatic\cxstatic.cpp 544

BYTE* pBuffer = new BYTE [ nBufferLen ];

...

delete pBuffer;

The programmer confuses delete and delete[]. This causes issues which may occur and may not. But you

should not do so anyway.

PVS-Studio: V519 The 'm_xSt' object is assigned val ues twice

Page 15: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

successively. Perhaps this is a mistake. resizedlg. cpp 244

m_xSt = CST_RESIZE;

m_xSt = CST_RESIZE;

Judging by the code, there must be m_ySt here. But we cannot keep from using copy-paste again and

again, right?

V531 It is odd that a sizeof() operator is multipli ed by sizeof().

pphtmldrawer.cpp 258

DWORD dwLen = ::LoadString(hInstDll, dwID, szTemp,

(sizeof(szTemp) * sizeof(TCHAR)));

There must be sizeof(szTemp) / sizeof(TCHAR) .

PVS-Studio: V556 The values of different enum types are compared:

enuHAlign == Center. cxstatic.cpp 151

if (enuHAlign == Center)

There must be enuHAlign == Midle. There is also if in the code nearby (enuVAlign == Middle) though it

must be Center. Confusion with enum, in short.

PVS-Studio: V501 There are identical sub-expression s to the left and

to the right of the '||' operator. resizedlg.cpp 15 7

HDWP CItemCtrl::OnSize(...)

{

...

if (m_styTop == CST_ZOOM ||

m_styTop == CST_ZOOM ||

m_styBottom == CST_DELTA_ZOOM ||

Page 16: Comparing the general static analysis in Visual Studio 2010 and PVS-Studio by examples of errors detected in five open source projects

m_styBottom == CST_DELTA_ZOOM)

...

}

Perhaps the code must look this way:

HDWP CItemCtrl::OnSize(...)

{

...

if (m_styTop == CST_ZOOM ||

m_styTop == CST_DELTA_ZOOM ||

m_styBottom == CST_ZOOM ||

m_styBottom == CST_DELTA_ZOOM)

...

}

Comparison results

We do not draw any certain conclusions. One of the tools was better in some projects and the other tool

was better in others.

References 1. Andrey Karpov, Evgeniy Ryzhkov. Difficulties of comparing code analyzers, or don't forget about

usability. http://www.viva64.com/en/a/0071/.