沒有QA？如何確保軟體開發品質

任何軟體專案或產品，達到高品質是開發團隊必然的目標。不過高品質並非垂手可得，它需要團隊的共識和努力才能達成。

確保品質有很多方式。過去常見的方式是瀑布式開發方式(waterfall)中，在程式設計師確定code complete之後，靠QA/QT/QC(註1)來執行測試，並且在測試週期中，驗證是否符合設計規格，並記錄追蹤問題(bug)，有時候甚且扮演催促修復的角色。因而，特別是大型團隊，專門「處理」品質的QA角色十分重要。很多時候，團隊可能面臨沒有QA的狀態，此時要如何確保品質呢？

為什麼會沒有QA

有時候，環境造就沒有QA的局面。例如，新創公司可能也只有5個人，無法有專責QA。又例如，大型企業中因資源分配不均，導致某些專案無法有專責QA。

但更多時候，沒有QA指的是，沒有能做「真正QA工作」的人。也許團隊裡面有許多人持有QA的職稱。但很可能僅做到QC/QT的工作(註1)。實務上，在軟體開發團隊中，實際做的事情其實比職稱來的重要。就品質的角度而言，QA大部分的時間應該花在開發循環「前期」或者「中期」。以Scrum中的Sprint來說，在kick-off時，QA應該花最多時間在定義DOD，決定產出的評斷標準，在sprint每天活動中，QA應該花時間在檢視產生的程式碼(code review)並且透過每個工作產出，主動改善現有品質。換言之，QA應該比單純的程式設計師，更會程式，更知道系統的交互作用細節，並能透過直接或間接修改程式，直接影響開發過程中的品質。因為，開發前中期的品質修正，效果好，成本低，遠比開發「後期」再來幾個測試循環來的有效！

簡要的說，能真正做QA工作的人，必須能比程式設計師團隊更會寫程式。起碼也要是「曾經」非常會寫程式。

如果團隊不巧沒有這樣的人，有三種方式可以在沒有QA的情況下確保軟體開發的品質：

方式一：Scrum

Scrum方法論中，概念上每個Scrum成員都是「同樣質量」。換言之，Scrum進行中重視的是產出，每個SPRINT的結果是「可交付的東西」。而Sprint中間要完成的細項，應該將品質涵蓋入內，而由自行取得該任務的人，完成其保證。

有許多作法和上面這段熬口的說明有關。首先，DOD (definition of done) 除了涵蓋unit-test之外，其實應該也涵蓋整合測試。如果不涵蓋整合測試，就應該另外有一個任務是專做測試。並且，每個story完成中，必定涵蓋這個story應該要有的使用測試(用以檢驗規格)以及回歸測試(用以檢驗是否有副作用)。這些測試，可以單獨成為一個工作，也可以作為DOD的一部分。

無論如何，基本概念是：「人人應該都可以生產程式碼，當然人人應該都可以測試」。實際執行時，或許有些人比較常「拿到」測試工作，但這並不代表這些成員就只是進行測試而已。有些人比較常拿到「寫程式性質」的工作，但並不代表這些人不負責品質。

Scrum的團隊重視每個Sprint的共同結果，此結構也讓沒有QA也能達到高品質。因此Sprint的長度不能太長，太長就會落入「團隊中自行區分QA和Engineer」的後果。

方式二：Pair Programming

Pair Programming是指兩個人一起用一台電腦，一個鍵盤來共同寫程式。這作法在2000年左右發展的Extreme Programming被大大推崇，不過能有決心推動的團隊並不常見。

由於Pair Programming讓每段程式碼至少都會被兩個人看過，而且在頭腦中想過。它可以避免大部分的低級錯誤(拼錯字)，也可以避免懶人錯誤(程式風格，漏寫unit-test)，然而，更重要的是它讓兩個程式設計師的真實想法，在執行同件事情的時候被「好好溝通」。而這更大幅避免對設計或需求的誤解。

Pair Programming似乎有效率和產能上的疑慮，但無論如何，它確實是在沒有QA的情況下，確保開發品質的絕妙方式。強烈建議閱讀一下wiki上的Pair Programming最下面的參考論文。

注意！

前兩個方式雖然符合敏捷開發的精神，並且能從系統結構層面，解決問題。然而，這兩種方式都必須要有結構性的改變，除非是剛剛成立的新團隊，要造成結構性的改變很困難，而且，即便做的好，也得花上其他心血才能有「能見度」，有能見度，才有所謂的功勞。

有兩個古時候的例子：

(a) 鶡冠子扁鵲：扁鵲曰「長兄最善，中兄次之，扁鵲最為下。」魏文侯曰：「可得聞邪？」扁鵲曰：「長兄於病視神，未有形而除之，故名不出於家。中兄治病，其在毫毛，故名不出於閭。若扁鵲者，鑱血脈，投毒藥，副肌膚，閒而名出聞於諸侯。」

(b) 孫子兵法：故善戰者之勝也，無智名，無勇功，故其戰勝不忒。

然而，對於一個軟體專案的主管而言，這些結構性的改變才是自己真正的價值。即便價值很難被衡量，但價值會永遠存在自己的手上。

如果短時間難以改變環境，可以考慮以下的方法三：

方式三：Part-time & Automation

工讀生(part-time student)和自動化測試(automation)似乎是兩個不同主題，但就確保軟體開發品質而言，把他們當做「一件事情」來處理，會有驚人的效果。

簡單的說，就是雇用3至4個優秀的工讀生，每週上班2-3天，組成工讀生團隊，執行測試任務，並且在熟練測試任務之後，開始進行測試自動化撰寫，並且在小組長(team leader)帶領下視情況參與更多品質管理的事情。這聽起來是個繁複的事情，但執行起來，遠比方法一二簡單。

其步驟如下：

(a) 選定一位以後想要朝專案經理或主管方向前進的優秀資深工程師，讓他作為工讀生團隊小組長

(b) 到各大學相關科系徵求大四以上的長期工讀生，一般來說，只要能妥善說明對他們未來就業的好處，通常可以找到足夠優秀的人。工讀生至少需要在職6個月以上。

(c) 組成團隊後，第一個月僅只需要熟悉目前軟體系統，第二個月才開始讓他們執行測試計畫，回報並記錄問題

(d) 在此過程中，由小組長指定測試內容和範圍，換言之，這段期間，其實小組長才是扮演QA的角色。而其他成員都可以將繁瑣的測試交給工讀生。然而，程式的品質仍然是所有成員負責，工讀生不在Scrum的範圍內，因此不「負擔責任」。

(e) 當測試進行2個以上SPRINT，工讀生應該已經開始覺得測試是很煩人的事情，但也應該知道品質對產品的重要性。這和在學校做專案計畫有天壤之別。因此，就可以開始由小組長領導工讀生進行測試自動化。

(f) 測試自動化並不期望把所有整合測試/回歸測試，100%統統自動化。只要把「簡單瑣碎」的測試自動化，通常就能節省一半以上的時間

(g) 通常六個月後，3-4人的工讀生團隊就能完成部分整合測試，和大部分的回歸測試。而下一輪的新工讀生，可以選擇從頭開始打造新的測試自動化，也可以接手前期工讀生做到一半的自動化。打造新的自動化通常可以用新的工具，新的角度來測試既存系統，可以讓品質在一次提高。接手前期工讀生的自動化，可以讓自動化範圍更廣，空出時間來做其他的事情


工讀生進行自動化測試的開發，對組織，對小組長，對工讀生有三贏的效果。(參考：實習生的三贏)

組織：讓沒有QA的團隊，能確保高品質的產出。除了要花些微的工讀費用之外，讓團隊成員能把瑣碎的事情下放給優秀的工讀生，使團隊成員能集中心力，但又同時負責個人生產的品質。同時，由於利用工讀生來培養小組長，讓組織能了解這個資深工程師，適不適合作為領導者，萬一不適合，頂多也是犧牲工讀生而已。

小組長：沒有人生下來就會當主管，當主管必須要有經驗，而工讀生團隊是主管最容易讓資深工程師測試自我的地方。因此小組長可透過這獨立運作的團隊，練習各種管理技能。

工讀生：大部分優秀的大四以上學生，都猜得到業界和學界的差異。許多人可能會在暑假應徵summer intern，然而其實短短兩個月，通常會做比較獨立的專案，雖然都很有趣，但是和在學校有很大的不同。加入實際開發團隊，即便只是做測試，也能了解「現實和學校」的差距，並且體會到軟體專案開發時，品質的重要性。讓有此六個月經驗的工讀生，更容易在未來找到更好的工作。

---

註1：關於Quality Control, Quality Test, Quality Assurance, test engineer, SQA的各種工作角色的區分，請參見wiki。然而，誠如前所述，工作角色名字不重要，做出事情才重要。

Self-complacency v.s. confidence

Recently, I've interviewed a candidate who applied for Sr. software engineer job and he did demonstrate an unusual character of Taiwanese engineer: self-complacency. 

The interview started from a quick technical review to know his skill set base on his resume.

At first, we quickly ask "How do you identify your Linux skills" and he said "I am very good at Linux". Then we ask a few basic Linux commands, for example: "How to know your current disk usage?". "How to know current Linux kernel version". He actually cannot answer much.

Then, we asked "How do you identify your level of SQL?" and again he said "I am very good at SQL". But he actually don't know any basic concept of "outer join" and "normalization". He explained SQL is what he is working on not the scope of his research area. Then we asked "how to avoid SQL injection in application" but he still can't say anything.

Since his resume said that he is "excellent" in English, we asked if we can continue interview in English. He took a deep breathe and then said "no".

Finally, he explained that he can quickly learn "anything" in one month but then forget that in next month. But he still emphasized that "I think I do have the talent of software development, give me a mission and I can commit the code". This might be true that he can learn "anything" quick and "commit code" for the job. But it is definitely also true that no one will hire such risky engineer.

To me, at first, I think it is extremely unusual to see an engineer with 7 years experience has such self-complacency and with such low actual skill set. But then his resume shows that he worked for 6 different companies in past 7 years. Change job doesn't means that the candidate is bad. But it is still a sign for Taiwanese programmer. Another sign is that he did have very good education background and also good delivery in first 2 jobs. But then, he seems stay in there. 

How to distinguish  self-complacency from self-confidence?

We do want engineers have confidence to take challenges. But how can we tell a candidate has confidence or just too complacency? 

During interview, our suggestions are

(1) Check skills by fact

Ask simple, un-ambiguous technical questions. Do not rely on his words on resume. For example, if his resume shows he has 10 years experience in Linux Operation, make sure you will ask at least 3 most simple questions and 5 middle level questions. If his resume says that he has 10 years in Java, make sure ask the last moment of his java coding. Sometime, the answer will surprise you.

(2) Check experience by real actions

Some candidates described experience like this "design XXXX system". This could be very important task but also could be a self-complacency thing in candidate's mind.

If this candidate lead design meeting, come out a design document - even a hand writing document or drawing picture from whiteboard. That means we can check the design and know how many of "special" things inside the design. 

However, if candidate couldn't say any concrete delivery (we don't need to ask him to provide right away, just want to know if there is), then it is very possible that he "believe" that he did the design job but actually the design was inside his mind. 

You need to heck the actual actions and delivery of it. You can ask "please let me know the exactly things you did on design XXXX system and also the delivery at that time"

(3) Find critical point by the actions candidates did

Always looking for fact of candidates did in previous jobs. It is very similar to (2). We will need to understand his behavior via his previous actions when face a critical situation.

However, I am sure everybody will prepare a pretty nice answer of "why you leave your current company". So just save your time, ask other question.

Check candidate's resume and you will always find some critical moment in his career. For example, a project which requires a new skill set, then you can ask how candidate learn new skills. Or, senior engineer might play leaders role in certain situation, then, you can ask what is the worst thing happened during his leading. 

Always Consider Facts

In short, always consider to gather fact during interview and identify if candidate fit your organization through fact, not to guess.

企業巫醫：公司需要你，還是你需要公司？

每年年底照例有不少人考慮轉換跑道

換工作是很個人的考量，每個人狀況有很大的差異。然而，當生涯成長變成首要因素時，其中一個考量點是：「公司需要你，還是你需要公司」？

盈利企業當然需要各式人才，讓企業持續成長獲利。但每個人在組織中的重要性不同，特別是上千人的大企業。即便主事者真心誠意認為「每個員工都是公司寶貴資產」，但就事實來看，上千人的組織，不可能「每個人」都是寶貴資產。

因此，在想要轉換跑道之前，先衡量一下，到底公司需要你的程度如何？請記得衡量標準不是你的個人能力，而是公司需要你的程度。

了解自己被需要的程度有幾個方向：

(1) 有機因素 vs 有毒因素

有些時候，某些人以為掌握特定資訊，是被需要的因素。然而，這是屬於有毒因素。換言之，如果只靠掌握某些資訊是很危險的(註1)

舉例來說：在大企業中，專案從開始到結束有一定的流程和系統需要專案經理維護資料的正確性，例如填寫各種表格之類。然而如果專案經理「很會」填表搜集資訊，並且甚至知道很多「秘訣」，這並不是真正「被需要的」理由，而僅只是資訊掌握。將這個因素，作為被需要的原因，就是屬於有毒因素。又或者，僅只是在一個組織中待了很久，知道許多「眉角」，這並非壞事，但僅靠這點不應該被依賴的原因。

大部分的人，都能靠自己的能力產出價值，這就是有機因素。最簡單的情況就是：一個程式設計師撰寫所需要的程式，自然就產出價值。又或者一個廚師的手藝驚人，自然可以產出有價值的菜色。

有機因素和有毒因素都可以被培養，但是，培養有機因素的道路比較長遠。

(2) 換！

有些工作，天生取代性高。例如：餐飲業的員工，只要有心學習，絕大部分的人都能成為一個好的餐飲業員工。

有些工作，很具有專業性，只是在某個特定組織內，取代性很高。例如：在大型軟體組織(例如微軟)的工程師，在數萬個工程師中，其實任一工程師並沒有絕對的必要性。

屬於這類型的狀態下，要提昇被需要的程度，最快的方式其實就是「換」。

但是在「換」之前，需要確保兩件事情：

(a) 現在的工作產出已經超乎別人的預期：也就是說，別人(同事,老闆)很確定你做的非常好了 

(b) 提昇自己的個人能力：說來簡單做起來難，尤其是在大組織中。但就現有的工作範圍往外擴張是最好的方式。

(3) 避免帕金森法則

帕金森法則：請先參考wiki的說明。

當在大企業做了3-5年之後，很有可能落入帕金森法則陷阱之中。也就是會盡量「擴大本來不需要的時間與資源用來完成工作」。這在過去的研究中有相當多的證實，並且全世界所有政府機關都有此現象。

一旦不自覺的落入帕金森法則，會下意識的降低自己的效率，產出和學習新技術的能力。自己的低效率，會成為成長的障礙。在員工能自我解決此問題之前，對大組織來說，都是「員工需要這個公司」而非「這個公司需要員工」。同樣的，對於小組織而言，公司需要員工的而非員工需要公司的機率會大得多。

---

註1：但是業務掌握客戶關係則不屬於此類，客戶關係並不是資訊，而是透過長期互動累積而來。是屬於資產的一種。
參考資料

參考資料
(1) 自我感覺良好的能力不足

How to build a self-motivated software developer team

A team which all team members are self-motivated is a dream team where everybody want to join in. But, it is really rare. Well, maybe the 1992 dream team was one of the cases but we didn't see many examples in software developer's world.

Nevertheless, to build a self-motivated team is a major task of a manager. If you are the manager, this should be your priority-zero task, and should be only one zero, if you are going to build a new team.

Self-motivated might be the best way to drive software engineer's performance. Since high productive engineer works high involved in creativities and mental work.  These are hardly be gain or impacted by salary or physical award (reference: 1, 2, 3). Sometimes it could be driven by a noble target: save all poor children in the world, increase average human life to 200 years. But unfortunately, that kind of things normally happens in non-profit organization with a born charismatic leader. 

So how to build a self-motivated software developer team in a common working place?

Three key points: (1) selection (2) intersection (3) automation.

(1) Selection

If a member who is already a self-motivated person, then it will be easy to keep his good attitude. Since it is pretty hard to change a person's mindset, therefore to get a right person is the most easy way.

Do select a person who already has record on self-motivation. DON'T select a person who has good skill-set but has bad records on self-motivation and you believe you can change his/her mindset, I won't say it is possible but you need to understand that you really have much less changes. Since you can't be with the person 7x24 and his mindset is only inside his brain.

The key point of selection is try to understand his past. Do not judge his "self-motivation" by ask future question like: will you be self-motivated in our team? Understand his past working experience by asking the difficult moment. Especially, sometime he might not be in spotlight and what he did on that moment.

Again, identify a person's past is the best way to select right attitude candidates. And a best fitting person can really have positive impact for the team. 

(2) Intersection

There is no way to read team member's mind. People have left-hand-column, even if you already selected best people to join your team and you were very sure they all are honest. But they surly have things which can't tell or won't tell.

Some leaders or managers will try to dig it out, it is fine to do but not that easy. The best way is to admit the existence. And then leaders should focus on the intersection which is the area that covers the target of a single team member, the team and the organization. And that is also the area where creates win-win-win.

A self-motived member could find that area by himself. But if you are leader/manager, you should keep in mind that figure out the area and knowing the area is very different in each individual. To fine tune a bit earlier can largely reduce the risk of un-motivated situation.

For example, "Ken" is always in good performance and deliver pretty good result. And you believe that you knew that his career interesting is pure technical area. Which means he might want to be an architect in the future. However, if you see he is pretty happy to lead junior members to work out teamwork activities, for example : organize a study group, handle company travel, or year end party. That might be a sign of that he is actually want to be a manager in the future. If his actions is toward his goal then it is fine. However, if his actions didn't reflect your assumption, then it is your time to reset yourself or/and make a little changes.

The bottom line is to avoid lose-lose-lose situation. This sounds funny and should not even considered to happen. But there are too many true sad stories existed in the world. An usual/typical sample in software developer's world was: a team (which members are well-pay) worked over-time on an ambiguous project for months, delivered a product which didn't fit PM's expectation, company lose money, members lose life, customer didn't get result: no body wins.

To avoid 3-lose situation, make sure there is an intersection of all. If you were the manager, make sure you can guide your members into his own intersection. But if you really see that there is NO intersection at all, make sure to create one.

(3) Autopilot

Autopilot means a member knows exactly what to do, how to do. And most important thing is if there are tasks out of his current skill set, he will know how to learn. He can even do self-consideration of intersection and selection. Totally worry-free.

Again, if the selection was done good, to let team member become autopilot will be easy. All you need to do is coaching. Coaching is another topic of a few books plus lots of trainings/experience. If you are an experienced leader/manager then you already know how to do, just reminder to keep open your mind. And also please keep in your mind that open-your-mind is never easy.

The bottom line is to avoid micro-management.  Software development is highly brain consuming job. Micro-management never works in software development, it does work in other kind of job but it definitely doesn't work in software development. If you need to talk to your member about when to arrive in office, when to leave, what to update in Slack, how to write email. That means you are NOT trying to build self-motivated team or this specific person won't fit in a self-motivated team.

Sorry, No short-cut

There is no short-cut in for organization theory. However, practically do the right things normally will get the positive result in a few months. Of course, sometimes in a rapid startup environment, it seems no time to wait. But most of us in software developer world are running marathon not a sprint.

Skill sets and performance concerns:

There is one more thing. I didn't discuss skill sets and performance here. But this is actually extremely critical for a software developer's team. Without an excellent skill sets and performance delivery the self-motivated will be useless. However, this "should" be not difficult to identify during interview and resume check.

Scrum: Feature-boxing of Sprints

In certain scenario, a scrum project could allow different length of Sprints and still satisfies Timeboxing principle. It benefits an involving software product of build MVP, increase integration quality and make members even more focus.



Time-boxing is one of fundamental practices in Scrum methodology. Most of actions in scrum are time-boxed. It makes sure the project keep going and also lets members focus on current task.

Time-boxing is one of the cure of Parkinson's Law of Triviality for all kind of meeting. No matter it is Sprint kick-off or daily standup. I believe it should apply to not only project management but also other area of organization, as long as the timeframe is a consensus of members. 

You can easily find the definition or discussion of how important time-boxing (or timebox) is in google. For example: this one or this one.

Timeboxing in Sprint

A Sprint in scrum also apply to time-boxing. Normally it means when project kick-off, all scrum members will decide how long is a Sprint. The recommended length is 4 weeks in software development. After that all Sprint will be in 4 weeks, no matter how many stories have done or what changes happened between Sprints. That also means if a project is in Sprint-11, 40 weeks was gone. 

A fixed timebox sprint make process happens. But is it necessary to keep the same length of Sprint?

Still timeboxing but flexible in next Sprint

Timeboxing is still critical and should be applied in the moment before event start. Sometimes, we should not fix timeframe in project kick-off and keep for next 40 weeks or more.

In fact, every Sprint kick-off meeting should also discuss a timeframe of this specific Sprint.

Image a sprint-kick-off meeting. Everybody: Product Owner, Scrum Master, Scrum Members were together in this Sprint kick-off meeting, discuss stories and story-points. If PO select 3 stories: P1, P2 and P3 which needs 2 weeks, 1 weeks, 3 weeks. What team should do? The general practice should be keep P1 and P2 in this Sprint and also select a few lower priority task to fill up that one left week. However, there are a few drawbacks in this situation. Firstly, remember Parkinson's Law? members might extends the time for not select lower priority tasks which sometimes is not a bad idea. But secondly, if a scrum team keep doing select a lower priority to fill up time slot, it make the "priority" not "priority" any more. 

If all members have consensus of flexible Sprint length, they can easily setup next Sprint to 3 weeks. And that Sprint will done  P1+P2. Or with Product Owner's agreement, they can setup next Sprint to 6 weeks to complete all P1~P3. 

In some scenario, the most beautiful way should be keep P1 in next sprint only. And this Sprint will be just 2 weeks. 

Make MVP possible

Why do one story at one Sprint is a beautiful way? Well, in some scenario, we need to make MVP(Minimum Viable Product)...and keep make MVP for a while to identify market and make sure minimum waste. Unfortunately, no body can foresee the future. A quick/early delivery is the reason why MVP is useful and also the reason why it is good to stick a single story delivery in a single Sprint. To achieve that, we need flexibility in different Sprint.

The key person of MVP is Product Owner (PO). And to make MVP possible, the bottom line of scrum is to have PO's real participant. In reality, it is very very hard to PO to focus on what he needs to do. A single story delivery forces PO to make a decision of the real priority of backlogs. Since the team will really delivery a things in every Sprint and of course PO should then put the thing in the market. PO can't say: well, it is too early to go for market. Since this is his priority one! Priority one should go for market right away! 

Why Quality improved?

Very simple, each Sprint with only one story has its own QA process and due to the complexity of change is short. we can easily fix integration issue or new defeat much quicker than a sprint with more than one story.

Agile: Feel comfortable to Adopt reality

An agile team should be comfortable to adopt changes and those changes should reflect to reality. But should still keep the critical concept. There is always a way to enjoy both without compromise to an uneasy method. 

To be flexible in Sprint length doesn't mean to break Timeboxing. But it does break some rule from the regular Scrum training. Nevertheless, process is continuously improving is the bottom and spirit of Agile. Scrum master or organization leader should take responsibility to make sure the process could be tailored after retrospective.

Case Study

As developer manager of a enterprise software product, I modify a bit of product release method to make sure PO's backlog is a real priority to reflect market needs. 

The result is since 2016/July to 2017/Mar (8 Month), we delivery 5 releases. The longest Sprint is 8 weeks and the shortest Sprint is 2 weeks. We actually complete 5 major stories and achieve following technical result.

(1) Quality: no side effect in these 5 release. Compare to previous fix Sprint length model much lower technical defeat in beta.

(2) Productivity: averagely, more code conduct per engineer compare to previous fix Sprint length model. However, at the same period of time, we enjoy more team building activity then before and very very very few overtime.

(3) Motivation: engineer could see the result (no matter good or bad) few weeks after delivery. Previous fix Sprint length model needs 3 to 9 months to know if our works really matters.

If you'd like to know more detail, please drop me email.